The impact of girl child marriage on the completion of the first cycle of secondary education in Zimbabwe: A propensity score analysis.

BACKGROUND: The association between girl child marriage and education is widely acknowledged; however, there is no large body of demographic studies from Zimbabwe that have addressed this aspect. This study aimed to examine the extent to which child marriage affects one academic milestone, i.e. completion of the Ordinary Level, the first cycle of high school, which is also the most critical indicator of educational achievement in Zimbabwe.
METHODS: We used the 2015 Zimbabwe Demographic and Health Survey and extracted 2380 cases of ever-married women aged between 20-29 years. We applied a propensity score-based method, which allowed us to mimic a hypothetical experiment and estimate outcomes between treated and untreated subjects.
RESULTS: Our results suggest that child age at first marriage is concentrated between the ages of 15-22, with the typical age at first marriage being 18 years. Both logistic regression and PSM models revealed that early marriage decreased the chances of completing the first cycle of high school. Regression adjustment produced an estimate of prevalence ratio (PR) of 0.446 (95% CI: 0.374-0.532), while PSM resulted in an estimate (PR = 0.381; 95% CI: 0.298-0.488).
CONCLUSION: These results have implications for Zimbabwe's development policy and suggest that girl-child marriage is a significant barrier to educational attainment. If not addressed, the country will most likely fail to meet sustainable development Goal 4.2 and 5.3. Social change interventions that target adults and counter beliefs about adolescent sexuality and prepubescent marriage should be put in place. Moreover, interventions that keep teenage girls in school beyond the first cycle of high school should be prioritised.

Introduction

Although many countries globally have ratified the Convention of the Rights of the Child’s [1] definition that any human being below the age of 18 is a child, girl child marriage is still a common practice in many parts of the world, especially in South Asia and sub-Saharan Africa [2]. Globally, an estimated 39 000 child marriages occur every day [3], and if these trends persist, approximately150 million girls will be married off by 2030, many against their will [4]. Statistics from sub-Saharan Africa indicate that the proportion of women who marry before their 18th birthday ranges from 19% in Namibia to more than 70% in Chad, Mali, and Niger [5]. It is also estimated that nearly 50% of the world’s child brides live in South Asia, while India alone accounts for about 30% of the global total [2]. Even in developed countries such as the United States of America, the statistics suggest that approximately 78400 children have been married between 2010 and 2014, with immigrant children being more likely to be married than their peers born in the United States [6].

Like in most other parts of Africa, Zimbabwe also has a high prevalence of girl child marriage throughout its 10 provinces, despite the awareness campaigns held in the past few decades [7]. For instance, recent data from the Multiple Indicators Survey [8] estimates that the country’s child marriage rate is 33%, slightly higher than the global average of 29%. Consequently, Zimbabwe is classified among the 41 countries globally with an alarming rate of child marriages [9, 10].

These statistics on child marriage are not only alarming but also suggest that the practice continues unabated. This is especially concerning because many such marriages often go unreported or unregistered, especially in rural areas [11]. Therefore, the exact scope of the problem might still be unknown, suggesting that child marriage might be a hidden and unaddressed problem [12].

Several reasons, most of which are steeped in the family institution, culture and the toxic combination between poverty and gender discrimination, have facilitated the continual exploitation of the girl child [12-15]. Some scholars have noted that poor parents are often compelled to marry off their daughters when faced with austerity, thus reducing their expenses as they will have one less person to feed, clothe and educate [16]. In some communities, the onset of menarche is considered the threshold for adulthood and a sign of marriage readiness [14, 17]. Also, some believe that getting a girl married early might have a protective factor against early sexual debut, loss of virginity, factors which if not protected, will affect the family’s honour [18].

In Zimbabwe, there are three main drivers of child marriage: cultural and (forced) marriage practices; religion; and, lack of policy enforcement. Regarding the latter, Hallfors et al. [19] note that while Zimbabwe’s constitution prohibits child marriages, the enforcement part is lacking. Forced marriage practices have prevailed despite the country being a signatory to the African Charter on the Rights and Welfare of the Child [20], which places an obligation on member states to end harmful practices such as child marriage, put in place effective actions, including monitoring the progress towards eradicating such practices [21]. In other words, prohibiting child marriage by law alone has had little effect in eliminating the practice [19, 22], perhaps due to “countervailing norms, or widespread exceptions” [23], some of which are discussed below.

Outside of policy, retrogressive marriage practices such as kuripa ngozi (virgin pledging), kutizira (unplanned pregnant marriages) and kuzvarira (pledged marriages) are also continually used to justify marrying off young girls in Zimbabwe [24]. In kuripa ngozi, young girls are married off to appease an avenging spirit, while kuzvarira is often a survival tactic where low-income families negotiate with wealthy families to marry off their daughters at a younger age in exchange for grains, cows or money. Kutizira is a practice where if a girl gets pregnant out of wedlock, she is expected to elope to her boyfriend or the person responsible for the pregnancy. This is done to mitigate the shame of premarital sex and childbearing out of wedlock while preserving the family honour.

The prevalence of child marriage in Zimbabwe has also been linked to religion. While the practice is not peculiar to one religious group, Chamisa et al. [25] note that girl child marriages are more prevalent among the apostolic religion adherents, especially the Johane Marange and Johane Masowe groups. These religious groups exploit young girls through their religious teachings and practices. For instance, it is common for young girls to be forced into marrying older men under the pretext that church leaders have been directed by the ‘holy spirit’ [19]. Those who fail to adhere to these ‘prophecies’ are threatened and cursed, forcing them to do certain things, even against their will. This practice has also endured due to the strong political ties between the Government of Zimbabwe and the Apostolic Faith groups, which commands a considerable following estimated to be about 34% of the country’s population [26]. As a result, Zimbabwean politicians have targeted this religious sect, using its pulpit to garner electoral support [19]. Critics have argued that these religious-political ties perhaps explain why policymakers tend to overlook child marriage issues among the Apostolic Faith sect as they fear losing a significant proportion of the electorate [19, 25].

Whatever the reasons for marrying off young girls, child marriage is a human right violation, with far-reaching consequences. Child marriage robs young girls of their childhood and is a significant setback for development [12]. It hinders their full participation in society as well as efforts to achieve gender equality broadly [7, 21]. Early marriage also confers risks to sexually transmitted diseases (STIs), including HIV and AIDS, early childbearing and the associated health hazards [27]. There is also evidence that child brides are more likely to become domestic violence victims and grow up feeling disempowered [15].

Undoubtedly, the demographic and health costs of early marriage are high, and governments across the globe cannot afford to allow this practice to continue. It is not surprising that several national and global initiatives have been put in place to find solutions to the problem. Child marriage is now a major policy issue and included in sustainable development goal no 5 [13, 28]. Global organisations and development partners such as UNICEF, UNFPA and Girls not Brides have also been working with many governments to improve the status of the girl child. Specific to Zimbabwe, platforms such as Girl Child Network and Childline have been used to raise awareness of the extent of the problem; however, progress remains slow and uneven [25].

Among the different solutions that have been proposed, education is considered the most important protective factor against girl child marriage [12, 13]. This is not surprising given that the returns to education, particularly secondary level education, are well documented [12, 13, 29]. Scholars have argued that lack of education curtails young girls’ full realisation of their rights, including the right to object to forced marriages and limiting their livelihood options [30]. However, the relationship between child marriage and education is not as straightforward, as it is influenced by several interconnected processes. For instance, some young girls may be forced to drop out of school due to poverty or academic ability, and, in a patriarchal society, marrying them off might be seen as the next rational step. At the same time, girls who are married early might fail to continue with their education due to different factors such as the financial situation, caring for a baby or simply because the husband might not allow them to [13, 31]. Put differently, child marriage could be a driver or a consequence of low educational attainment. To tease out the possible association with educational attainment, there is a need for methods that can isolate the true effect of child marriage from any confounding. However, there is not a large body of demographic studies that have addressed this aspect [13, 31].

To put Zimbabwe into context, studies on child marriage, in general, are limited. Where these have been done, the focus has been on determinants or prevalence [7, 25, 32, 33], legal and development focused efforts to end the practice [9, 34] or the association with reproductive health [19, 35]. Moreover, most of these studies on child marriage have been small scale, based on -non-probability sampling and limited to geographic areas, with population-based studies [32] being undertaken infrequently. Notably lacking is empirical research that has focused on the association between child marriage and educational outcomes, particularly secondary level schooling, which is the focus of the present study. Therefore, this study aims to address these gaps by examining the impact of early marriage on completing the first cycle of secondary school. We also draw on a propensity score-based approach that allows us to measure the effect of early marriage on educational attainment.

Materials and methods

Data

The data used in this study came from the 2015 Zimbabwe Demographic and Health Survey (ZDHS), a cross-sectional nationally representative study of health and demographic indicators of women aged 15–49. Mindful that our focus was on completion of the first cycle of secondary school (11th year of schooling), which cannot be achieved before the age of 17 years if one is on time, we restricted our analysis to 2380 ever married or partnered women who were aged 20–29. Thus, we gave at least three more years for the youngest cohort to complete this level of schooling. We also excluded women who were not in their first union as they would not have provided information related to their first spouse. This is particularly important in cases where the first union was before the woman’s 18th birthday.

The 2015 ZDHS, was undertaken by the Zimbabwe National Statistics Agency (ZIMSTAT) in collaboration with the Ministry of Health and Child Care (MoHCC) and the United Nations Population Fund (UNFPA). Data were collected from approximately 11 000 households which were sampled using a two-stage cluster sampling design with two strata (urban and rural) for each province. At the first stage, 400 enumeration areas delineated by the 2012 Zimbabwe Population Census sampling frame were selected, of which 166 were urban areas and 234 rural areas. The second stage involved the selection of individuals at the household level.

Measures

The outcome variable was the completion of lower secondary school, while early marriage was the treatment variable. We present the descriptions of the variables as used in this study below:

Outcome

Lower secondary school completion (yes = 1, no = 0), indicated whether an individual had completed the first cycle of high school or not. In the ZDHS, respondents were asked questions about the highest grade or years of schooling they had completed by the survey date. We considered individuals who had completed 11 years of schooling as having completed lower secondary education. Zimbabwe has a 7-4-2 basic education structure consisting of seven years of primary, followed by two cycles of secondary level education; four years of General Certificate of Education, also called Ordinary level and two years of General Certificate of Education at the Advanced Level [36]. The Ordinary level, which is the 11th year of schooling, is generally used to determine student achievement, progression to either A-level, polytechnic or teachers’ colleges, as well as employment status. While learners who want to progress to university might opt to pursue the Advanced Level Education, the majority, due to several factors such as quality of pass, lack of fees or career choice, often end at the Ordinary level. Thus, completion of this level is the most critical indicator of educational achievement, such that a person who holds the Ordinary level certificate is referred to informally as having ‘finished school’. In consequence, most young people ‘finish school’ as early as 17 years old. Given the fragile economic landscape in Zimbabwe, characterised by high unemployment, closure of companies, retrenchments and poverty, marriage often becomes the next best option for young people who are not in school.

Treatment variable

Our main explanatory or treatment variable was early marriage, a binary indicator based on age at marriage (<18 years, and ≥18 years).

Other variables of interest

We also included the following factors that were found in the previous research to be associated with early marriage and developmental outcomes such as education and coded them as follows:

Spousal age difference has been identified as a proxy for female autonomy in reproductive health research. Using data on the woman and her partner’s actual age, which was captured as a continuous variable in the ZDHS, we created the spousal age difference variable by subtracting the wife’s age from that of her partner.

Polygamous marriage: This was a dummy variable to distinguish between (0 = monogamous marriage and, 1 = polygamous marriage).

Husband’s education: 1 = primary level education or lower; 2 = secondary level education, 3 = higher education.

Union characteristics

Sexual debut

Early sexual debut is often associated with teenage pregnancy, and both factors have a reciprocal effect on both early marriage and educational attainment [12]. At the same time, early marriage can also determine sexual debut for many young girls, by extension, placing the latter on the causal pathway between child marriage and educational attainment. Mindful of this, we opted to distinguish between women who had their sexual debut before marriage and those whose age at first sex was after marriage. Hence, we use this variable as one of the controls and coded this as a binary variable to; (1 = before marriage; 2 = after marriage).

Sex of the household head

We also included the sex of the household head (0 = male; 1 = female) as a control. Female-headed households are increasingly becoming common in Zimbabwe, and research evidence suggests a higher incidence of poverty in these households relative to male-headed ones [37]. As such, it is often suspected that female-headed households have limited resources to invest in their children’s education, with girls being the most affected.

Sociodemographic controls

Given the literature which suggests that a nexus between religious beliefs and child marriage [25] as well as the prevalence of the latter in rural areas [24], we also included the following covariates: the place of residence (1 = rural, 2 = urban); and religion (1 = christian, 2 = apostolic, 3 = other).

Analytical method

We first explored the associations between the treatment variable (a binary indicator of early marriage) and potential confounders. Since analyses were survey-weighted, we used the Rao-Scott X2 test for the categorical variables and the weighted t-test for the continuous variables (Table 1).

Table 1

Descriptive statistics (% and means) for child marriage, and associations with selected variables of married women aged 20–29, 2015 ZDHS, n = 2380.

Variable	Early marriage	Late Marriage	Raw data
	N = 821 (36.8%)	N = 1559 (63.2%)	P-value	ASMD	Matched ASMD
Place of residence			<0.001
Rural	76.2	55.4		0.528	0.018
Religion			<0.001
Christian	35.5	54.9		0.441	0.021
Apostolic	54.8	39.9		0.346	0.009
Others	9.7	5.2		-	-
Husband education (years)			<0.001
0–10	31.3	19.3		0.215	0.009
11+	23.9	33.8		0.04	0.011
Missing	44.8	46.8		-	-
Type of marital union			<0.001
Monogamy	73.2	82.1		0.278	0.038
Polygamy	16.2	13.8		0.276	0.047
missing	10.6	4.1		-	-
Age at first sex			<0.001
Before marriage	17.3	41.7		0.586	0.002
Sex of household head			0.134
Female	34.0	37.6		0.125	0.008
Spousal age difference b	7.3±5.1	5.7±4.8	<0.001	0.344	0.028
Age of respondent (years) b	24.2±2.7	25.1±2.9	<0.001	0.318	0.058

Note: Reported percentages are population-weighted. We only report p-values for the original data and ASMDs for the matched data [43]. ASMD means absolute standardised mean difference and is a numeric value calculated for every covariate. An ASMD of <0.1 is generally taken to indicate a negligible difference between the treatment and the control group for that covariate [45, 46].

b mean ± standard deviation.

Given that our primary objective was to quantify the effect of early marriage, we opted for a methodology that controls for confounding factors. Therefore, we used propensity score matching (PSM) to estimate treatment effects and compared the results with those obtained from the conventional regression adjustment and the unadjusted estimate of the treatment effect. Though some methods of using the propensity score (PS) have been described in the statistical literature [38], we favoured PSM in this study due to its convenience in implementation and computation.

In this setting, we defined the PS as the conditional probability that a participant had an early marriage, conditional on the covariates. For the propensity score analysis, two steps were involved. First, we performed a binary logistic regression model of main effects by regressing the treatment variable (a binary indicator of early marriage) on all the covariates identified as potential confounders in Table 1 to generate propensity scores. We adopted the strategy of including as many covariates as possible based on previous research and scientific understanding [39, 40]. Second, we used the estimated propensity scores to create a matched sample based on the single nearest neighbour algorithm [41-43]. We implemented a 1–1 matching with replacement and without caliper.

We assessed the estimated propensity scores for sufficient overlap in the distribution of the treated and control groups (Fig 2). The PS adjustment objective is to create a matched sample in which the distribution of the covariates is the same between the treatment and control groups. We thus verified that covariate balance had been induced in the matched samples by examining their pre-matched and post-matched ASMDs [44]. While there is no consensus cut-off, ASMD values >10% may indicate covariate imbalance [45, 46].

Fig 2

The distributions of the estimated propensity scores for the early and late marriage participants before and after matching.

We then determined the effect of early marriage by estimating the prevalence of the outcome in treatment and control respondents separately in the matched sample. The prevalence estimation incorporated the survey weights to obtain population-level estimates [47]. The treatment effect was quantified as the prevalence ratio (PR) of early marriage respondents who completed the Ordinary Level of secondary schooling to the late marriage respondents who also completed the same level. For the regression adjustment, the traditional multivariable logistic regression was used to calculate the adjusted PR. However, the survey weights were accounted for in the analysis to account for the survey design. Further, where necessary, variance estimations accounted for the design of the survey. While we acknowledge the high likelihood of inducing bias, the missing indicator method [48] was applied to missing values on husband education to increase the sample size for the propensity score and multivariable models. We carried out all analyses in Stata v14. Where necessary, we reported two-sided p-values and p-values < 0.05 were considered statistically significant.

Results and discussion

Fig 1 shows the distribution of the sample by age at first marriage. We find that age at first marriage is concentrated between the ages of 15–22. The typical age at first marriage is 18, followed by 17 and 19 in that order.

Fig 1

Sample distribution of age at first cohabitation.

Table 1 shows the weighted descriptive statistics of the association between different variables and early marriage. We observed that the married respondents in their adolescence accounted for 36.8% of the total sample, while 63.2% had married after their 18th birthday. Table 1 shows that all our covariates (column 4) were significantly associated with the treatment variable, girl child marriage. There were statistically significant differences (p < 0.05) in all the confounders between the two groups, implying systematic differences in the risk factors between the two groups of interest.

Fig 2 is a density plot that shows the propensity scores distribution, both pre-matching and post-matching, for the two groups of age at marriage. As shown in the figure, compared to pre-matching, the propensity scores distribution was almost identical for the two groups after matching on the propensity score. After applying PSM, all the covariates were balanced, and there were no statistically significant differences (p > 0.05) between the two groups (results not shown but available on request). The ASMD values (Table 1, columns 5 and 6) most of the covariates were also substantially under the 0.10 threshold [43].

We do not present the results from the covariates’ estimates obtained in the first logistic regression (although these are available upon request). Both the unadjusted and all the adjusted estimates suggested that early marriage significantly (p < 0.05) associated with high school completion (Table 2). However, all the adjusted estimates increased the PR from the unadjusted estimate of 0.328 (95% CI: 0.285–0.378). After PSM adjustment, the prevalence of high school completion was 61.9% lower for those who had an early marriage (PR = 0.381; 95% CI: 0.298–0.488). Regression adjustment produced a PR estimate of 0.446 (95% CI: 0.374–0.532), indicating that the prevalence of high school completion was 55.4% lower for those who had an early marriage.

Table 2

Association of child marriage and completion of lower secondary school.

	Unadjusted	Regression-adjusted	PSM
Early marriage prevalence (%)	0.199	0.236	0.173
Late marriage prevalence (%)	0.609	0.529	0.454
PR (95% CI)	0.328 (0.285, 0.378)	0.446 (0.374, 0.532)	0.381 (0.298, 0.488)
P-value	<0.0001	<0.0001	<0.0001

Prevalence ratio (PR); CI: Confidence interval.

Sensitivity analysis

Finally, we performed a sensitivity analysis (Table 3) to examine the magnitude of hidden bias that would alter our estimated treatment effects and inferences. In other words, we determined how robust our matching analysis is to unobserved confounding variables. The details of the sensitivity analysis procedure are given elsewhere [41].

Table 3

Bias analysis of sensitivity to unmeasured confounding.

Gamma (Γ)	Q_mh+	Q_mh-	p_mh+	p_mh-
1	10.685	10.685	0.000	0.000
1.1	11.445	9.946	0.000	0.000
1.2	12.143	9.274	0.000	0.000
1.3	12.794	8.661	0.000	0.000
1.4	13.404	8.099	0.000	0.000
1.5	13.980	7.580	0.000	0.000
1.6	14.526	7.096	0.000	0.000
1.7	15.045	6.645	0.000	0.000
1.8	15.541	6.221	0.000	0.000
1.9	16.015	5.822	0.000	0.000
2	16.471	5.445	0.000	0.000
2.1	16.909	5.087	0.000	0.000
2.2	17.332	4.747	0.000	0.000
2.3	17.741	4.422	0.000	0.000
2.4	18.136	4.113	0.000	0.000
2.5	18.520	3.816	0.000	0.000
2.6	18.893	3.532	0.000	0.000
2.7	19.255	3.258	0.000	0.001
2.8	19.608	2.995	0.000	0.001
2.9	19.952	2.742	0.000	0.003
3	20.288	2.497	0.000	0.006
3.1	20.617	2.261	0.000	0.012
3.2	20.937	2.032	0.000	0.021
3.3	21.252	1.811	0.000	0.035
3.4	21.559	1.596	0.000	0.055
3.5	21.861	1.387	0.000	0.083

Γ: odds of differential assignment due to unobserved factors. Q+MH: Mantel-Haenszel statistic (assumption: overestimation of treatment effect). Q−MH: Mantel-Haenszel statistic (assumption: underestimation of treatment effect). p-_mh: significance level (assumption: overestimation of treatment effect). p+_mh: significance level (assumption: underestimation of treatment effect).

Under the no hidden bias assumption (Γ = 1), the QMH test-statistic indicates a significant treatment effect. Considering the bounds that assume we have over-estimated the true treatment effect (Q+MH), the treatment effect is significant under Γ = 1. It becomes even more significant for increasing values of Γ. However, for the bounds under the assumption of under-estimating the treatment effect (Q-MH), a value as high as Γ = 3.4 or more would be required for the results not to be significant at the 5% level. Therefore, our matching analysis suggests that early marriage is significantly associated with non-completion of high school education and is vulnerable to hidden bias. We should thus interpret the matching results with caution.

Discussion and conclusions

The current study investigated the effect of child marriage on high school completion. This study suggests that early marriage is still prevalent in Zimbabwe and perhaps a hidden crisis given the limited attention in empirical research. We found that among women in the age group 20–29, approximately 37% were married before their 18th birthday, placing Zimbabwe amongst the countries with the highest rates of underage married girls in Africa [9, 10]. This enduring practice continues to undermine all efforts towards gender equality. It must, therefore, be urgently addressed if the country is to make progress towards meeting sustainable development goal number five.

The fact that child marriage is associated with low educational attainment is well known [23, 29, 49, 50]; however, the exact extent of the impact is less clear given relatively few studies have measured this aspect, especially in Zimbabwe. In this study, we used propensity score matching [43] to statistically balance women who were married early and those who married late in terms of various factors that could potentially influence their likelihood of completing the first cycle of high school. Consistent with the few existing quasi-experimental studies on early marriage [13, 30, 51], our results suggest women who were married before their 18th birthday had a lower propensity to complete the first cycle of high school relative to their peers with similar characteristics but had married after their 18th birthday. We also found evidence of confounding. Before matching, women who married as children were 55.4% (regression adjusted) less likely to complete lower secondary school. However, the co-efficient increased to 61.9% after matching, suggesting that self-selection plays a part in explaining the relationship.

What exactly do these findings mean for Zimbabwe? First, the results suggest that girl-child marriage is a significant barrier to educational attainment. If the practice of child marriage is not addressed, Zimbabwe will most likely fail to achieve sustainable development goals 4.2 and 5.3 which seek to i) ensure that all boys and girls complete primary and secondary education and ii) “eliminate all harmful practices, such as child, early and forced marriage” respectively [28]. Considering the risks associated with early marriage [15, 19, 32], delaying early marriage should be at the forefront of development policy in Zimbabwe. Clearly, legislation alone has failed to curb the practice [19, 25]; thus, there is a need for the Government of Zimbabwe and its local and global development partners to intensify existing campaigns against child marriage. In particular, social change interventions that target adults and counter beliefs about adolescent sexuality and prepubescent marriages should be put in place. Similar programmes have been implemented in countries such as India and Ethiopia, with considerable success in reducing both the prevalence of child marriage and altering social norms [52]. Second, given the empirically determined link between child marriage and educational attainment, education plans in the country should integrate the goal of eradicating child marriage. Interventions that target girls who drop out of school must also be put in place, while existing programmes that provide school subsidies for vulnerable young girls should be broadened [27]. Further, interventions that help keep young girls in school beyond the lower secondary level might be beneficial to eradicating child marriage. Being out of school has been linked to risky sexual behaviours and unintended pregnancies- both of which are catalysts for early marriage [52].

While propensity score matching in this study has enabled us to examine the effect of early marriage completion of lower secondary school, some limitations should be considered when interpreting the findings. First, we acknowledge that the relationship between educational attainment and early marriage is complex as the latter can be both the cause and consequence of dropping out of school. Thus, there is a possibility of reverse causality between child marriage and educational attainment; however, PSM does not correct for this. Another limitation is that the DHS data does not include variables such as school quality or reasons for dropping out, which can shed more light on the temporal sequencing of child marriage and school leaving. In other words, more research is needed, which explores the effect of these factors on educational attainment so as to effectively tease out the effect of child marriage on secondary school completion. Finally, propensity score matching does not control for unobserved confounding. There may be other risk factors which we were not able to control.

Despite these limitations, our study makes significant contributions to the literature on child marriage in general and offers valuable insights into the phenomenon of child marriage in Zimbabwe. However, longitudinal and cohort studies are still needed to validate these findings as well as tease out causation.

24 Nov 2020

PONE-D-20-28233

The impact of girl child marriage on the completion of  the first cycle of secondary education in Zimbabwe: a propensity score analysis

PLOS ONE

Dear Dr. Bengesai,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

As you will see, the reviewers have comments on a wide range of issues, the most important of which concern the appropriate of the methods used, the presentation of the results, and the discussion of the results.

Please submit your revised manuscript by Jan 08 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

David Hotchkiss

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For more information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially sensitive information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: No

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The paper highlights the importance of addressing child marriage by exploring its impact on secondary school completion in Zimbabwe. While the authors have done a nice job of reviewing the evidence base on child marriage in Zimbabwe, unfortunately the arguments are not sufficiently supported by rigorous evidence in a manner that the current analysis has data to address. The survey of literature presents interesting cultural factors such as sexual initiation, virgin pledging and pregnancy triggered marriages. However, the contributions of these factors are not reflected in the analysis. While it is understandable that marriage characteristics data is not available, it should be possible to explore the contribution of pregnancy preceding marriage. This is an important missed opportunity. Thus the review of the literature raises many more questions than it answers. Overall, at a substantive level, is an important considerations is to address whether child marriage really the critical driver of schooling or or are other factors at play such as early sexual initiation, unintended pregnancy, school quality and the absence of services to address these. Without considering these other factors, and acknowledging that child marriage itself may be driven by them, it is misleading to conclude that more needs to be done to child marriage, particularly as it is not clear what precisely can be recommended as a strategy for addressing child marriage directly rather than some of its other drivers. A second important reservation I have about the paper is in terms of methods. With the use of retrospective data, and without considering the temporal sequencing of school dropout and child marriage it is misleading to conclude that child marriage is the driver when the reverse may also be true. There are indeed several papers (cited below) that call attention to the fact that at least part of the association may be related to underlying factors that lead to school dropout, which then leads to early marriage and/or pregnancy. This possibility of reverse causality has important implications for policy. Solutions to improving school outcomes may well need to focus on school quality and not child marriage per se. Finally, the paper includes sample weights as a variable in propensity score matching. This is certainly unusual and it is not obvious to the current reviewer how it is justified. That needs to be explained better.

Biddlecom, Ann; Gregory, Richard; Lloyd, Cynthia B.; Mensch, Barbara S. Associations between premarital sex and leaving school in four Sub-Saharan African countries. Studies in Family Planning. 2008; 39(4):337–350. [PubMed: 19248719]

Birdthistle, Isolde; Floyd, Sian; Nyagadza, Auxillia; Mudziwapasi, Netsai; Gregson, Simon; Glynn, Judith R. Is education the link between orphanhood and HIV/HSV-2 risk among female adolescents in urban Zimbabwe? Social Science & Medicine. 2009; 68(10):1810–1818. [PubMed: 19303688]

Case, Anne; Ardington, Cally. The impact of parental death on school outcomes: Longitudinal evidence from South Africa. Demography. 2006; 43(3):401–420. [PubMed: 17051820]

Lloyd, Cynthia B.; Mensch, Barbara S. Marriage and childbirth as factors in dropping out from school: An analysis of DHS data from sub-Saharan Africa. Population Studies. 2008; 62(1):1–13. [PubMed:

Reviewer #2: The authors estimate the effect of child marriage on the probability of completing the ordinary level of secondary school in Zimbabwe. They recognize that this estimate is likely to be confounded and attempt to control for measured confounding using propensity score matching. I appreciate that the authors recognize the fact that estimates of this effect from observational studies are subject to a great deal of bias and likely far from the true causal effect. However, the analytic methods used in this study need to be better explained and justified before it is suitable for publication. My detailed comments on each section of the manuscript follow.

Introduction

The introduction and framing of this argument could be improved. In particular, the information on line 118 regarding results from Koski and Heymann (2018) is partially incorrect. That study found that immigrant children living in the United States were more likely to be married than their peers who were born in the United States, but immigrants comprise a minority of the population of the United States and so it is incorrect to state that the majority of those married were immigrants. Please correct this.

Lines 139-141 include the following sentence: “In some communities, such as in Iran and Nigeria, the onset of menarche is considered the threshold for adulthood. Hence, girls who reach this biological threshold are perceived to be ready for marriage. (14, 17).” It is unlikely that all citizens of these two nations hold this belief and the authors should be more cautious in their interpretation. Reference number 17 is not sufficient to support this statement in Nigeria. It appears to include a single sentence that is not based on research and, unfortunately, it is published in a known predatory outlet, making it unclear whether the work was peer reviewed.

Data and statistical analyses

The paragraph that explains the educational structure in Zimbabwe (lines 260-271) is important. However, the authors do not address how well this structure corresponds with the measured variables available in DHS data, which typically include a continuous measure of the total number of years of schooling and a second categorical measure that indicates the highest level (no schooling, primary, secondary, or more than secondary) an individual attended or completed. (Notably, the 2015 DHS report for Zimbabwe includes the categories listed above and does not differentiate between completion of the ordinary and advanced levels of secondary school.) How did the authors use these variables to capture completion of the ordinary level of secondary school versus the advanced level? What is the expected magnitude of misclassification of educational attainment, and what effect would such misclassification be expected to have on their estimates?

The authors appear to confuse the concept of selection bias (line 308, line 439) with confounding, or may be using these terms interchangeably, which is confusing and should be corrected. (In epidemiology, all of the bias discussed in this paper would be categorized as confounding.) I agree with their assertion that estimates of the association between child marriage and educational attainment are very seriously confounded and could result from reverse causality. Hypothetically speaking, in order to identify the causal effect of child marriage on this outcome, one would need to compare a group of girls who married prior to the age of 18 with a group who married at the age of 18 years or later and those girls would need to be exactly the same in every way except for their age at the time of their first marriage. Clearly, it is very difficult to attain such a comparison in an observational study based on DHS data. In reality, girls who marry before the age of 18 are very different from those who marry at later ages in many ways. Unfortunately, the most important confounders of this relationship, such as childhood socioeconomic conditions and attitudes toward gender equality among the girls’ family and community prior to her marriage, are not captured in the DHS. This severely limits the extent to which confounding can be controlled, even through use of propensity score matching.

The authors use propensity score matching in an effort to make the treatment groups more exchangeable with regard to a small number of measured variables, but the rationale for the inclusion and treatment of these variables and the omission of others is unclear. For example, it is unclear why the respondent’s age and the age difference between spouses were categorized. Why not use age and age difference as continuous variables to improve the granularity of matching (and therefore further limit confounding)? Also concerning is that any couples in which the wife was older than her husband would yield a negative number, and negative numbers of any magnitude appear to be grouped in the category �0-4 years. Are the authors sure that all women in their sample were in their first marriage? If not, the age difference between the woman and her second (or later) partner may not be relevant to this analysis. Why was ethnicity not included in the propensity score model? It is one of the few pre-exposure variables available to the authors and is strongly correlated with child marriage in many countries, though I am not familiar with how this is distributed in Zimbabwe.

The authors cite work by Donald Rubin, a pioneer in the development of propensity score methods, but in some cases do so erroneously. For example, Rubin does not advocate for throwing any and all variables into the propensity score model as suggested on lines 320-321; in particular, including variables that may be a consequence of the exposure (in this case, marriage before the age of 18) is typically inappropriate. This is where use of cross-sectional surveys such as the DHS becomes even more problematic. Most variables in the DHS are measured in adulthood, well after marriage, and may be influenced by that marriage. For example, the categorical measure of household-level socioeconomic status included in all DHS is based on characteristics at the time the survey was conducted. One can easily imagine that child marriage may be a means for parents to secure places for their daughters in households with a higher standard of living; the authors acknowledge this in the introduction to their paper. This makes it inappropriate to adjust for this variable. The same argument could be made for place of residence measured after marriage.

The specific determination of matches based on propensity scores is also insufficiently explained. For example, the authors state that they used nearest neighbor matching, but more information is needed. Were calipers used to define a maximum acceptable difference in scores to assign a match? Were controls sampled with replacement? How many observations were off support, meaning that no reasonable match could be found?

The authors include estimates of the absolute standardized differences in their results. Why are values for these differences missing for some values in Table 1?

The authors refer to estimates of “heterogeneity” in the treatment effect (line 237, lines 470-472), but no such analyses were conducted as far as I can tell. The estimates in Table 2 appear to be for the entire sample and no discussion of tests for heterogeneity across sup-groups is included.

Given that data from this survey is cross-sectional and measures prevalence, I recommend that the authors report prevalence differences and prevalence ratios rather than odds ratios. Including absolute measures of effect (i.e. prevalence differences) would strengthen the paper substantially. This can be done after running a logistic regression model by using the – margins – command in Stata.

Discussion

The term “quasi-experimental” (line 432) typically refers to natural experiments in which exposure is plausibly random; it is inappropriate in this study in which exposure is far from random and a large degree of unmeasured and residual confounding likely remain in the estimates.

Lines 473-475 seem to indicate a fundamental misunderstanding of the DHS. The survey collects data from a nationally representative sample of individuals; the data are at the individual-level, not at the national-level.

I find the authors’ discussion of the limitations of their research cursory and insufficient. For example, “…our data was drawn from self-reported responses; thus, there is a possibility of underreporting of sensitive information.” What information used in this study would they categorize as sensitive and why? Would such underreporting be likely to bias their estimates? They note that propensity score matching does not control for unmeasured confounding but say nothing about the very high likelihood that their results remain substantially confounded even after accounting for the variables included in their propensity score model. Quantitative bias analysis to estimate how robust their estimates are to the presence of unmeasured and residual confounding would improve the paper immensely.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: Yes: Sajeda Amin

Reviewer #2: Yes: Alissa Koski

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

29 Jan 2021

Response to Reviewers

Dear Editor,

We would like to thank the reviewers for their comments as well as the opportunity to revise our manuscript for further consideration in your journal. We have addressed all the comments from the reviewers and below we provide a detailed description of the revisions made. These changes are also highlighted in yellow in the revised manuscript we submitted.

Reviewer #1

The paper highlights the importance of addressing child marriage by exploring its impact on secondary school completion in Zimbabwe. While the authors have done a nice job of reviewing the evidence base on child marriage in Zimbabwe, unfortunately the arguments are not sufficiently supported by rigorous evidence in a manner that the current analysis has data to address. The survey of literature presents interesting cultural factors such as sexual initiation, virgin pledging and pregnancy triggered marriages. However, the contributions of these factors are not reflected in the analysis. While it is understandable that marriage characteristics data is not available, it should be possible to explore the contribution of pregnancy preceding marriage. This is an important missed opportunity. Thus the review of the literature raises many more questions than it answers. Overall, at a substantive level, is an important consideration is to address whether child marriage really the critical driver of schooling or or are other factors at play such as early sexual initiation, unintended pregnancy, school quality and the absence of services to address these. Without considering these other factors, and acknowledging that child marriage itself may be driven by them, it is misleading to conclude that more needs to be done to child marriage, particularly as it is not clear what precisely can be recommended as a strategy for addressing child marriage directly rather than some of its other drivers.

Response: We concur that early sexual debut and teenage pregnancy are reciprocal to early marriage as well as educational attainment. Therefore, we included age at sexual debut as well as age at first birth as controls.

A second important reservation I have about the paper is in terms of methods. With the use of retrospective data, and without considering the temporal sequencing of school dropout and child marriage it is misleading to conclude that child marriage is the driver when the reverse may also be true. There are indeed several papers (cited below) that call attention to the fact that at least part of the association may be related to underlying factors that lead to school dropout, which then leads to early marriage and/or pregnancy.

Response: The DHS does not include any variables that measure school type or quality. We acknowledge this as a limitation in our study.

This possibility of reverse causality has important implications for policy. Solutions to improving school outcomes may well need to focus on school quality and not child marriage per se.

Response: While we acknowledge the relevance of school quality on educational attainment, we cannot make conclusions on this as school quality was not measured in the ZDHS. We acknowledge this as a limitation.

Finally, the paper includes sample weights as a variable in propensity score matching. This is certainly unusual and it is not obvious to the current reviewer how it is justified. That needs to be explained better.

Response: Though it is correct to say that it's not a usual approach, including survey weights in the propensity score model as an additional covariate is one of the recommended methods for estimating population-level treatment effects (see references 1,2 below). However, for simplicity sake, we have modified the propensity score matching analyses to accommodate survey weights when estimating the effect of treatment, but not when estimating the propensity score model (as suggested by reference 3 below).

Austin PC, Jembere N, Chiu M. Propensity score matching and complex surveys. Statistical methods in medical research. 2018 Apr;27(4):1240-57.

DuGoff EH, Schuler M, Stuart, EA. Generalising observational study results: applying propensity score methods to complex surveys. Health Serv Res. 2014; 49(1): 284-303. DOI: 10.1111/1475-6773.12090

Zanutto EL. A comparison of propensity score and linear regression analysis of complex survey data. Journal of data Science. 2006 Jan 1;4(1):67-91.

Reviewer #2: The authors estimate the effect of child marriage on the probability of completing the ordinary level of secondary school in Zimbabwe. They recognize that this estimate is likely to be confounded and attempt to control for measured confounding using propensity score matching. I appreciate that the authors recognize the fact that estimates of this effect from observational studies are subject to a great deal of bias and likely far from the true causal effect. However, the analytic methods used in this study need to be better explained and justified before it is suitable for publication. My detailed comments on each section of the manuscript follow.

Introduction

The introduction and framing of this argument could be improved. In particular, the information on line 118 regarding results from Koski and Heymann (2018) is partially incorrect. That study found that immigrant children living in the United States were more likely to be married than their peers who were born in the United States, but immigrants comprise a minority of the population of the United States and so it is incorrect to state that the majority of those married were immigrants. Please correct this.

Response: Corrected as suggested.

Lines 139-141 include the following sentence: “In some communities, such as in Iran and Nigeria, the onset of menarche is considered the threshold for adulthood. Hence, girls who reach this biological threshold are perceived to be ready for marriage. (14, 17).” It is unlikely that all citizens of these two nations hold this belief and the authors should be more cautious in their interpretation. Reference number 17 is not sufficient to support this statement in Nigeria. It appears to include a single sentence that is not based on research and, unfortunately, it is published in a known predatory outlet, making it unclear whether the work was peer reviewed.

Response: Changed to: In some communities, the onset of menarche is considered the threshold for adulthood and sign of readiness for marriage.

Reference also changed to: Raj A, Ghule M, Nair S, Saggurti N, Balaiah D, Silverman JG. Age at menarche, education, and child marriage among young wives in rural Maharashtra, India. Int J Gynaecol Obstet. 2015

Data and statistical analyses

The paragraph that explains the educational structure in Zimbabwe (lines 260-271) is important. However, the authors do not address how well this structure corresponds with the measured variables available in DHS data, which typically include a continuous measure of the total number of years of schooling and a second categorical measure that indicates the highest level (no schooling, primary, secondary, or more than secondary) an individual attended or completed. (Notably, the 2015 DHS report for Zimbabwe includes the categories listed above and does not differentiate between completion of the ordinary and advanced levels of secondary school.) How did the authors use these variables to capture completion of the ordinary level of secondary school versus the advanced level? What is the expected magnitude of misclassification of educational attainment, and what effect would such misclassification be expected to have on their estimates?

Response: We used information on years of schooling and deduced that those who completed 11 years of schooling would have completed the Ordinary level of Education, based on the 7-4-2 education structure in Zimbabwe.

The authors appear to confuse the concept of selection bias (line 308, line 439) with confounding, or may be using these terms interchangeably, which is confusing and should be corrected. (In epidemiology, all of the bias discussed in this paper would be categorized as confounding.) I agree with their assertion that estimates of the association between child marriage and educational attainment are very seriously confounded and could result from reverse causality. Hypothetically speaking, in order to identify the causal effect of child marriage on this outcome, one would need to compare a group of girls who married prior to the age of 18 with a group who married at the age of 18 years or later and those girls would need to be exactly the same in every way except for their age at the time of their first marriage. Clearly, it is very difficult to attain such a comparison in an observational study based on DHS data. In reality, girls who marry before the age of 18 are very different from those who marry at later ages in many ways. Unfortunately, the most important confounders of this relationship, such as childhood socioeconomic conditions and attitudes toward gender equality among the girls’ family and community prior to her marriage, are not captured in the DHS. This severely limits the extent to which confounding can be controlled, even through use of propensity score matching.

Response: Replace selection bias with confounding

Response: done as suggested

The authors use propensity score matching in an effort to make the treatment groups more exchangeable with regard to a small number of measured variables, but the rationale for the inclusion and treatment of these variables and the omission of others is unclear. For example, it is unclear why the respondent’s age and the age difference between spouses were categorized. Why not use age and age difference as continuous variables to improve the granularity of matching (and therefore further limit confounding)? Also concerning is that any couples in which the wife was older than her husband would yield a negative number, and negative numbers of any magnitude appear to be grouped in the category �0-4 years.

Response: As suggested, we have used age as a continuous variable. However, we chose to leave age difference as categorical to make it easier to use the missing indicator method to deal with its missing values, which are quite many.

Are the authors sure that all women in their sample were in their first marriage? If not, the age difference between the woman and her second (or later) partner may not be relevant to this analysis.

Response: We included the variable union, to control for whether the woman was in the first or subsequent union.

Why was ethnicity not included in the propensity score model? It is one of the few pre-exposure variables available to the authors and is strongly correlated with child marriage in many countries, though I am not familiar with how this is distributed in Zimbabwe.

Response: We did not include ethnicity as this was not measured in the 2015 ZDHS

The authors cite work by Donald Rubin, a pioneer in the development of propensity score methods, but in some cases do so erroneously. For example, Rubin does not advocate for throwing any and all variables into the propensity score model as suggested on lines 320-321; in particular, including variables that may be a consequence of the exposure (in this case, marriage before the age of 18) is typically inappropriate. This is where use of cross-sectional surveys such as the DHS becomes even more problematic. Most variables in the DHS are measured in adulthood, well after marriage, and may be influenced by that marriage. For example, the categorical measure of household-level socioeconomic status included in all DHS is based on characteristics at the time the survey was conducted. One can easily imagine that child marriage may be a means for parents to secure places for their daughters in households with a higher standard of living; the authors acknowledge this in the introduction to their paper. This makes it inappropriate to adjust for this variable. The same argument could be made for place of residence measured after marriage.

Response: We have only included variables that affect either the exposure or outcome, not those that may be a consequence of the exposure.

We also acknowledge the possibility of measurement error, given the ZDHS only provides current household wealth information, and not prior to the marriage. However, it is not unreasonable to assume that young women are likely to marry within the same socio-economic status (12). As well, it is plausible that those who marry in wealthy families might be allowed to continue with their education post marriage. Taking this view, we contend that estimations which include household wealth are superior to those without, as they give us some indication of the effect of socio-economic status, albeit with certain limitations.

The specific determination of matches based on propensity scores is also insufficiently explained. For example, the authors state that they used nearest neighbor matching, but more information is needed. Were calipers used to define a maximum acceptable difference in scores to assign a match? Were controls sampled with replacement? How many observations were off support, meaning that no reasonable match could be found?

Response: Done as suggested. More details about the propensity score matching have been included.

The authors include estimates of the absolute standardized differences in their results. Why are values for these differences missing for some values in Table 1?

Response: the values missing have now been included appropriately

The authors refer to estimates of “heterogeneity” in the treatment effect (line 237, lines 470-472), but no such analyses were conducted as far as I can tell. The estimates in Table 2 appear to be for the entire sample and no discussion of tests for heterogeneity across sup-groups is included.

Response: references to heterogeneity removed

Given that data from this survey is cross-sectional and measures prevalence, I recommend that the authors report prevalence differences and prevalence ratios rather than odds ratios. Including absolute measures of effect (i.e. prevalence differences) would strengthen the paper substantially. This can be done after running a logistic regression model by using the – margins – command in Stata.

Response: done as suggested

Discussion

The term “quasi-experimental” (line 432) typically refers to natural experiments in which exposure is plausibly random; it is inappropriate in this study in which exposure is far from random and a large degree of unmeasured and residual confounding likely remain in the estimates.

Response: Corrected as suggested.

Lines 473-475 seem to indicate a fundamental misunderstanding of the DHS. The survey collects data from a nationally representative sample of individuals; the data are at the individual-level, not at the national-level.

Response: This limitation has been removed.

I find the authors’ discussion of the limitations of their research cursory and insufficient. For example, “…our data was drawn from self-reported responses; thus, there is a possibility of underreporting of sensitive information.” What information used in this study would they categorize as sensitive and why? Would such underreporting be likely to bias their estimates? They note that propensity score matching does not control for unmeasured confounding but say nothing about the very high likelihood that their results remain substantially confounded even after accounting for the variables included in their propensity score model. Quantitative bias analysis to estimate how robust their estimates are to the presence of unmeasured and residual confounding would improve the paper immensely.

Response: The limitations have been revised as follows:

While the use of propensity score matching in this study has enabled us to examine the effect of early marriage completion of lower secondary school, some limitations should be considered when interpreting the findings. First, we acknowledge that the relationship between educational attainment and early marriage is not straightforward as the latter can be both the cause and consequence of dropping out of school Thus, there is a possibility of reverse causality between child marriage and educational attainment; however, PSM does not correct for this. Another limitation is that the DHS data does not include variables such as quality of school or reasons for dropping out, which can shed more light on the temporal sequencing of child marriage and school leaving. More research is needed which explores the effect of these factors on educational attainment so as to effectively tease out the effect of child marriage on secondary school completion. Finally, propensity score matching does not control for unobserved confounding. There may be other risk factors which we were not able to control.

Sincerely,

Dr Annah Bengesai

University of KwaZulu-Natal

South Africa

Email: bengesai@ukzn.ac.za

ORCID 0000-0002-2711-8530

Dr Lateef A. Babatunde

University of Ilorin, Nigeria

Email: Amusa.lb@unilorin.edu.ng

And

Dr Felix Makhonye

University of KwaZulu-Natal

Email: MakonyeF@ukzn.ac.za

Submitted filename: Response to Reviewers_PONE.docx

Click here for additional data file.

18 Feb 2021

PONE-D-20-28233R1

The impact of girl child marriage on the completion of the first cycle of secondary education in Zimbabwe: a propensity score analysis

PLOS ONE

Dear Dr. Bengesai,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

As you will see, one of the reviewers has a number of comments on how you addressed the comments on the initial version of the manuscript.  Please address these comments an provide point-by-point responses.

Please submit your revised manuscript by Apr 04 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

We look forward to receiving your revised manuscript.

Kind regards,

David Hotchkiss

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: No

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: No

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: The revised manuscript is much improved with regard to the description of how the outcome variable was measured and the inclusion of a sensitivity analysis. However, the fundamental problems with confounder identification and adjustment remain and must be addressed.

In my initial review I recommended that the authors measure age and age difference variables continuously to improve the accuracy of matching. This was done for the wife’s age but not for the age difference between spouses. I don’t follow the logic of using spousal age difference a categorical variable because there were a large proportion of missing values. Missing values for wife’s age and/or husband’s age would prohibit the estimation of age difference regardless of whether the variable was measured continuously or categorically. Moreover, it is unclear how missing values were handled in the analysis. At a later point in the paper (lines 371-374) the authors state, “The missing indicator method was applied to missing values on covariates such as spousal age difference…” but it is unclear what this means and no citations are provided. If this refers to the use of binary variables to indicate missingness, the authors should acknowledge the high likelihood that this method itself induces bias. (For reference, see Donders et al. Journal of Clinical Epidemiology 2006.)

In my initial review I raised concerns regarding control for variables that may be affected by the exposure, including household socioeconomic status (SES). The authors assert that it is reasonable to assume that young women marry husbands of the same SES, suggesting that SES is the same prior to and after marriage. This contradicts much of what is known about the relationship between poverty and child marriage. It is also contradicted by the introduction to the paper, which indicates that “…kuzvarira is often a survival tactic where low-income families negotiate with wealthy families to marry off their daughters at a younger age in exchange for grains, cows or money.” This certainly suggests that SES differs prior to and after marriage, especially in the case of child marriage, and should not be adjusted for. For further information on identifying potential confounders, I recommend that the authors consult recent work by VanderWeele (Principles of confounder selection, European Journal of Epidemiology, 2019).

In my initial review I asked for clarification regarding whether all women were in their first unions. Given that the DHS only collects information on a woman’s current spouse, women in second or later unions would not have provided relevant information about their first spouse. In response, the authors included a variable for union number in the model. Union number cannot possibly affect age at marriage and therefore cannot be a confounder. It should not be adjusted for. An alternative would be to exclude women who were not in their first unions from the analysis, though this may result in problems with selection bias.

In the revised version of the manuscript the authors have added age at first sex and age at first birth to their model. While sex and pregnancy may lead to child marriage, for a substantial number of young girls, age at marriage probably determines age at first sex and age at first birth. For girls whose first birth follows their marriage, this variable is almost certainly on the causal pathway between age at marriage and educational attainment. Again, this strongly indicates that these variables should not be treated as confounders. The proportion of girls who report an age at first sex prior to age at first marriage (which can be estimated using DHS data) might guide subsequent analytic decisions.

On line 463 the authors report the prevalence of child marriage among women between 17 and 24 years of age. This doesn’t correspond with the analytic sample of 20-29-year-olds described in the rest of the paper. Is this a typing error?

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: Yes: Alissa Koski

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

5 Apr 2021

Dear Editor,

We would like to thank the reviewers for their comments as well as the opportunity to revise our manuscript for further consideration in your journal. We have addressed all the comments from the reviewers and below we provide a detailed description of the revisions made. These changes are also highlighted in yellow in the revised manuscript we submitted.

Reviewer #2: The revised manuscript is much improved with regard to the description of how the outcome variable was measured and the inclusion of a sensitivity analysis. However, the fundamental problems with confounder identification and adjustment remain and must be addressed.

In my initial review I recommended that the authors measure age and age difference variables continuously to improve the accuracy of matching. This was done for the wife’s age but not for the age difference between spouses. I don’t follow the logic of using spousal age difference a categorical variable because there were a large proportion of missing values. Missing values for wife’s age and/or husband’s age would prohibit the estimation of age difference regardless of whether the variable was measured continuously or categorically. Moreover, it is unclear how missing values were handled in the analysis. At a later point in the paper (lines 371-374) the authors state, “The missing indicator method was applied to missing values on covariates such as spousal age difference…” but it is unclear what this means and no citations are provided. If this refers to the use of binary variables to indicate missingness, the authors should acknowledge the high likelihood that this method itself induces bias. (For reference, see Donders et al. Journal of Clinical Epidemiology 2006.)

Response: we have used both age and spousal age difference as continuous variables

In my initial review I raised concerns regarding control for variables that may be affected by the exposure, including household socioeconomic status (SES). The authors assert that it is reasonable to assume that young women marry husbands of the same SES, suggesting that SES is the same prior to and after marriage. This contradicts much of what is known about the relationship between poverty and child marriage. It is also contradicted by the introduction to the paper, which indicates that “…kuzvarira is often a survival tactic where low-income families negotiate with wealthy families to marry off their daughters at a younger age in exchange for grains, cows or money.” This certainly suggests that SES differs prior to and after marriage, especially in the case of child marriage, and should not be adjusted for. For further information on identifying potential confounders, I recommend that the authors consult recent work by VanderWeele (Principles of confounder selection, European Journal of Epidemiology, 2019).

Response: we have removed wealth quintile from the analysis.

In my initial review I asked for clarification regarding whether all women were in their first unions. Given that the DHS only collects information on a woman’s current spouse, women in second or later unions would not have provided relevant information about their first spouse. In response, the authors included a variable for union number in the model. Union number cannot possibly affect age at marriage and therefore cannot be a confounder. It should not be adjusted for. An alternative would be to exclude women who were not in their first unions from the analysis, though this may result in problems with selection bias.

Response: we have excluded women who were not in the first union from the analysis

In the revised version of the manuscript the authors have added age at first sex and age at first birth to their model. While sex and pregnancy may lead to child marriage, for a substantial number of young girls, age at marriage probably determines age at first sex and age at first birth. For girls whose first birth follows their marriage, this variable is almost certainly on the causal pathway between age at marriage and educational attainment. Again, this strongly indicates that these variables should not be treated as confounders. The proportion of girls who report an age at first sex prior to age at first marriage (which can be estimated using DHS data) might guide subsequent analytic decisions.

We have differentiated between women who had sexual debut before marriage and those whose debut was after marriage

On line 463 the authors report the prevalence of child marriage among women between 17 and 24 years of age. This doesn’t correspond with the analytic sample of 20-29-year-olds described in the rest of the paper. Is this a typing error?

We have corrected the error

We have also copy edited the manuscript for grammatical errors.

Submitted filename: Response to Reviewer_PONErev2.docx

Click here for additional data file.

17 May 2021

The impact of girl child marriage on the completion of the first cycle of secondary education in Zimbabwe: a propensity score analysis

PONE-D-20-28233R2

Dear Dr. Bengesai,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

David Hotchkiss

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: (No Response)

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: Yes: Alissa Koski

21 May 2021

PONE-D-20-28233R2

The impact of girl child marriage on the completion of the first cycle of secondary education in Zimbabwe: a propensity score analysis

Dear Dr. Bengesai:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. David Hotchkiss

Academic Editor

PLOS ONE