Understanding children’s harmful work: a review of the methodological landscape

Children’s engagement with work has been widely researched using a wide variety of methods. However, the extent to which such methods and their combination provides insight into forms of children’s harmful work (CHW) is not obvious. This paper reviews and assesses respective opportunities and challenges of the main methods that have been used to study children’s engagement with work. It proposes research design principles and a methodological landscape for an integrated approach to child-centred, inclusive, and ethical research of CHW.


About ACHA:
The research informing this Working Paper as well as its publication was made possible thanks to the Foreign, Commonwealth & Development Office (FCDO)-funded research on Action on Children's Harmful Work in African Agriculture (ACHA). The aim of the programme is to build evidence on: • the forms, drivers, and experiences of children's harmful work in African agriculture; and • interventions that are effective in preventing harm that arises in the course of children's work.
It is currently assumed that the majority of children's work in Africa is within the agricultural sector. However, the evidence base is very poor in regard to: the prevalence of children's harmful work in African agriculture; the distribution of children's harmful work across different agricultural value chains, farming systems and agro-ecologies; the effects of different types of value chains and models of value chain coordination on the prevalence of harmful children's work; and the efficacy of different interventions to address harmful children's work. These are the areas that ACHA will address.
ACHA is a collaborative programme led by the Institute of Development Studies (IDS), Brighton, UK. Partners include: ACHA is directed by Professor Rachel Sabates-Wheeler (r.sabates-wheeler@ids.ac.uk) and Dr James Sumberg.
We are grateful to Carolina Szyp for her invaluable support to this paper by providing background research and streamlining all references.

References 23
Figures and tables  Table 1 Overview of national household surveys and measurement of child labour 7 Table 2 Opportunities and challenges of methods to gain insight into prevalence of forms of CHW 14 Table 3 Opportunities and challenges of methods to gain insight into drivers and dynamics of forms of CHW 15 Table 4 Opportunities and challenges of methods to gain insight into the impact of (interventions on) forms of CHW 17 Table 5 Opportunities and challenges of mixed-methods design to gain insight into forms of CHW 19 Contents 1 Introduction 1 Some countries also adopt their own definitions of hazardous child labour, such as Côte d'Ivoire.
Forms of children's harmful work (CHW) are notoriously difficult to identify, assess and understand. Common definitions of child labour such as 'worst forms of child labour' and 'hazardous child labour', as put forward by the International Labour Organization (ILO), are premised on notions of hazard and risk 1 but do not include an explicit consideration of harm (Maconachie, Howard and Bock 2020). Harm can be considered 'an identifiable negative impact on an individual or household arising from a specific workplace hazard' (ibid: 8) and CHW 'refers to any work that children undertake that actually results in harm to the child and/or their household' (Sabates-Wheeler and Sumberg 2020: 8). Forms of CHW are often hidden from sight and its prevalence, drivers and impacts are highly context specific (Maconachie et al. 2020). Research on CHW therefore requires careful consideration of its methodological approach and individual methods. This paper provides a review of methods that are commonly used for studying child labour and children's engagement with work (sections 1-4) in order to input into research design of the Action on Children's Harmful Work in African Agriculture (ACHA) programme (section 5). In doing so, it explores the opportunities and challenges of common methods and proposes an innovative methodological landscape for studying CHW. This review is guided by several principles. First, we consider methods that are used for studies across the spectrum of child labour and children's work. However, in keeping with the focus of ACHA, we reflect specifically on how those methods are used for studying children's engagement with work that may be considered hazardous or harmful. Second, we adopt an interdisciplinary approach, considering methods that are used across the social sciences, including anthropology, childhood studies, economics and geography, among others. Third, we pay special attention to child-centred research. Methodologies that adopt a child-centred approach typically try to understand different types of harm in relation to emic notions of childhood. Studies without such an approach tend to pay little attention to the nuances of work, view children as victims of circumstances or ignorance, and push for abolishing children's work through a wholesale ban. Fourth, we consider the extent to which methods are inclusive and incorporate views and voices across identities and groups, notably gender, age, disability, religion and faith, and ethnicity.
We review three types of methods: survey methods; qualitative and participatory methods; and certification methods. Survey methods range from nationally representative multi-purpose household surveys to purposive child labour surveys that are administered to smaller populations. Generally, surveys collect information with relatively large thematic and population coverage but with relatively limited participant involvement. They are rarely administered directly to children.
Qualitative and participatory methods range from focus group discussions and individual interviews to participant observation and visual methods. These methods tend to involve the research population more actively than surveys do, although levels of involvement depend on the specific method. Although qualitative methods are commonly applied to smaller samples, it is important to note that some participatory methods themselves can operate well at scale. These include individually oriented and group-based activities such as drawing, structured visuals, qualitative interviews and focus groups carried out with larger groups.
Certification methods are tools that are used within certification schemes (Ton et al. 2020). Firms use certification to reduce reputational risk in relation to children's work within their supply chains. To check compliance, data on production and household livelihoods are collected by producers and auditors, usually with little engagement with children or their caregivers. While methods within such mechanisms are akin to surveys and qualitative methods, we review them separately as their use is prescribed as part of voluntary standards, and data are collected indirectly (through producers). The inability to directly feed into processes of research design and implementation means that these methods are therefore less flexible compared to primary academic research.
We also review mixed methods as an overall approach to research design. Studies that have adopted mixed-methods research designs explicitly seek to achieve both breadth and depth by combining a variety of methods. Methods could be mixed in parallel or sequentially. Mixing methods leads to both larger analytical coverage and greater population involvement.
Inevitably this framing presents an oversimplification of the range of methods and their many intricacies. Furthermore, many studies adopt a combination of methods and data, often in implicit ways without making reference to a mixed-methods approach (such as using different qualitative and participatory tools in small-scale studies). This categorisation serves as a framework for organising this review as opposed to a strict delineation.
The remainder of this paper is structured as follows. First, we provide an overview of methods as outlined above, exploring their use within studies of child labour and children's work. Second, we assess the merits and challenges of specific methods for assessing the prevalence of forms of children's harmful work, drivers and dynamics, and impact (in line with ACHA's core research foci). Finally, we propose research design principles and a methodological landscape for studying CHW, aiming to inform research design for the ACHA programme.

Review of methods
This section provides an overview of methods using the categorisation as listed above.

Survey methods
A wide range of survey methods exist for studying children's engagement with work, ranging from large-scale surveys that collect information about work alongside many other topics, to purposive small-scale and child-centred surveys. We explore some of the most common survey methods below.

National multi-purpose household surveys
National multi-purpose household surveys collect information across a range of issues and are representative at country level. Living Standards Measurement Studies (LSMS), Multiple Indicator Cluster Surveys (MICS), Demographic and Health Surveys (DHS) and Labour Force Surveys (LFS) have been widely used to gain insights into the prevalence and patterns of child labour at a national level (Bhalotra and Tzannatos 2003;International Programme on the Elimination of Child Labour (ILO/IPEC) and Statistical Information and Monitoring Programme on Child Labour (SIMPOC) 2007; Understanding Children's Work 2017). These surveys often do not produce detailed information on child labour but collect information on employment of household members, characteristics of the household and its members, and wider household living standards, which can help to understand the context in which child labour takes place (Verma 2008 In turn, the International Conference of Labour Statisticians (ICLS) translates these conventions in statistical terms and sets standards for measurement of child labour (ibid.).
The narrow focus of these conventions and their rigid standards result in a similarly narrow remit in most multi-purpose surveys. Nevertheless, surveys differ in their potential to explore children's engagement with work. Within an LSMS, for example, the ability to cross reference information about children's work with data on school attendance and educational attainment, as well as demographic and socioeconomic characteristics of the household and its members, contributed to the popularity of the LSMS for studying child labour (Bhalotra and Tzannatos 2003). An MICS provides insights into children's engagement with unpaid household chores, which are not captured in many other surveys (Dayıoğlu 2013). A notable downside of the MICS is that information about health and nutrition is only collected for children under five years of age, which limits the ability to link information about children's engagement in work to health and nutrition outcomes (ILO/IPEC-SIMPOC 2007). Similarly, use of the DHS is limited due to a small range of questions about employment and only asking these to individuals aged 15-49 years. LFS is the most comprehensive in terms of capturing information about employment, but age brackets vary across surveys, with lower age thresholds included ranging from 10-15 years (Desiere and Costa 2019). Table 1 provides a comparative overview of national household surveys and their potential use for studying child labour.

Child labour surveys
Child labour surveys include a wide set of purposively developed survey instruments, ranging from large-scale household-based surveys to small-scale surveys with street children (Verma 2008). The Statistical Information and Monitoring Programme on Child Labour (SIMPOC) (also the statistics and monitoring unit) of the ILO's International Programme on the Elimination of Child Labour (IPEC) has played a key role in developing survey-based instruments and in advising national governments on how to generate high-quality data on child labour (SIMPOC n.d.).
Household-based child labour surveys use the household or family unit as an entry point into understanding patterns of child labour, with parents/guardians and children acting as respondents (ibid.). SIMPOC has developed questionnaires and methodologies in support of National Child Labour Surveys (NCLSs). Questionnaires can be implemented as a standalone survey or be attached to other surveys (SIMPOC/ ILO n.d.), such as LFS. Questionnaires commonly consist of three parts: (1) household roster; (2) adult questionnaire; and (3) child questionnaire (children aged 5-17) (ILO 2017). Given the purposive nature of an NCLS, they provide detailed information about child labour, certainly in comparison to multi-purpose household surveys. For example, as it includes children aged 5 and upwards, it allows for assessing the age at which children started working (ILO 2015). The questionnaires do not capture engagement in domestic chores or unpaid care work and therefore do not provide a full representation of children's engagement with work, particularly for girls, who are more likely to be engaged in housework.
Child-focused surveys include children and/or youth as respondents. A well-established survey is the School-to-Work Transition Survey (SWTS), which aims to gain better insights into transitions from school into work, and to understand transitions into the labour market for youth (Elder 2009). The survey is directed at youth aged 15-29 years, and its underlying sampling methodology aims for national representation. Although it is possible to use SWTS for producing child labour estimates, its main objective is to supplement the information collected through LFS or NCLS and provide detailed data about the supply of youth labour (ibid.).
Another category of child-focused surveys includes those that are developed and implemented as part of specific research studies. These vary widely in scope, sampling and types of questions asked. Examples include a six-country study that assessed whether child domestic work can be considered as a worst form of child labour, which administered questionnaires to more than 3,000 children aged 6-18 years (Gamlin et al. 2015) and a study of work and education in slum settlements in Dhaka among 2,700 children aged 6-14 years (Quattri and Watkins 2016). School-based surveys collect information about how work affects school attendance or performance and attitudes to schooling (SIMPOC n.d.). Schools are used as primary sampling units with questionnaires being administered to children in those schools. Surveys may also include interviews with teachers, administrative staff and parents, and may also include a control group of children out of school (Verma 2008). While other surveys generally limit questions to school enrolment and attendance, school-based surveys seek to generate data about how much time children spend in school, how often they miss school because of work, and their ability to engage in homework and extracurricular activities (Guarcello, Lyon and Rosati 2005). Large-scale school-based surveys were undertaken in the early 2000s with support from ILO/IPEC in Brazil, Kenya, Lebanon, Sri Lanka and Turkey, among other countries (ibid.).
Establishment surveys focus on the demand side of child labour and collect information from employers or labour intermediaries. Surveys seek to interrogate the situation in the workplace with questions focusing on the nature of work, hours of work, remuneration and pay, injuries and illnesses, and other conditions of work. Establishment surveys are rarely representative of all establishments employing children as identification of such establishments is inherently problematic (Verma 2008). However, they use the place of employment as an entry point, so they can be valuable for collecting information about forms of labour for children that live outside of the household unit or at non-registered locations, such as children living on the streets (ILO/ Special Action Programme to Combat Forced Labour (SAP-FL)/ IPEC 2012).

Impact evaluation surveys
Impact evaluation represents a growing body of research within which surveys are used to collate information about children's engagement with work. They often employ multi-purpose surveys with varying degrees of detail on children's work, typically based on the examples reviewed above. While evaluations of programmes that seek to reduce child labour as a primary objective tend to include more detail about children's engagement with work, this is less often the case for evaluations of interventions in which reducing child labour is a secondary objective. The policy area of social protection is a case in point.
Social protection has become a key policy area for reducing child labour (ILO 2018). Subsequently, an increasing number of studies consider the impact of social protection programmes -including schemes such as unconditional cash transfers, conditional cash transfers and public works programmes -on children's engagement in work (Dammert et al. 2018;de Hoop and Rosati 2014).
In the large majority of cases, evaluations aim to capture the programme effects on an array of outcomes, and child labour tends to be only one such outcome, resulting in relatively narrow collection of information.

Small-scale surveys
The use of survey methods is not limited to collection of large-scale data. Qualitative researchers also use small-scale quantitative surveys to develop their knowledge of the research setting, introduce themselves, and to get specific data that are important to their analysis of children's lifeworld, work, education and social position (Dyson 2014a;Hashim 2004;Katz 2004;Reynolds 1991).
In her research on child labour in the Zambezi Valley, Reynolds conducted a census of 12 families in her research setting (Reynolds 1991). She had already worked in the community before commencing the study and thus had a broad knowledge of it. By contrast, in her study in south-eastern Sudan, Katz saw her village-wide household survey as a way to introduce herself and her research, while constructing a socioeconomic and cultural profile of the community. The survey illuminated the diversity of economic activities people were engaged in, both on-and off-farm, and their seasonality (Katz 2004). In the context of a child-centred study on everyday involvement in rural household labour in a remote village in the high Himalayas in Nepal, Dyson undertook a full village census on age, educational background and occupation of all household members (Dyson 2014a).

Qualitative and participatory methods
Qualitative studies span a range of scales, from small case studies zooming in on a limited number of people to large-scale studies working with samples of several hundreds. Some methods are used successfully with a limited input of time, while others require a substantial investment of time to develop dense relations within the community and to capture the culture and context (Lancy 2015: 387). Wessells, on the basis of his work with child soldiers in East Africa (interviewed in Johnson and Lewin, forthcoming), has suggested that 'mini-ethnographies' should be carried out as part of any research seeking to understand child protection in community settings. This is particularly relevant in gaining insights into CHW that may at first be invisible, and in understanding how intergenerational relationships play out in workrelated family, community and employer/employee practices and decision making. Adequate time also needs to be allowed for adults to accept the participation of children and to support them to be part of the research (Chawla and Johnson 2004).
A wide range of methods is available within the qualitative and participatory toolbox. They are rarely used in isolation but rather in combinations that serve to elucidate different aspects of a research question. Increasingly, more traditional methods such as interviews and observations are used alongside creative methods (Boyden and Ennew 1997;Mitchell 2006;Punch 2001b). The sequencing of methods is flexible and depends on whether the aim is to map a set of factors that can be explored in depth or at scale later in the research, or to unpack processes surrounding children's work. Participant observation is a key element of ethnographic studies throughout the time spent in the research setting. In early phases of the research, observations are broad, focusing on grasping the general organisation of everyday life, including the work that children do. Later in the research, observations become gradually more focused on specific aspects of children's lives. Observations can involve random observation of each child in the sample throughout an entire day to discover the range of activities they engage in, accompanying children (and adults) to learn from them and participate in their work, and talking to children and adults for many hours (Dyson 2014;Johnson, Hill and Ivan-Smith 1995;Katz 2004;Punch 2001a;Reynolds 1991). The time spent informally with various participants gives them time to engage with the research. In her study, Punch (2001b) questions the extent to which adult researchers can do participant observation with children. She argues that there are limits to participation because although researchers can join children's games and work, the researcher will always be a different type of player in the game (ibid: 165; Atkinson 2019).

Participant and other types of observation
Time-use studies and diaries to map children's work are time consuming but one of the best ways of getting insights into the multiple tasks that children undertake during a day and their ability to combine different chores, work and play. Children's work can be recorded in different ways such as random 'snapshots' of labour allocation, 24-hour reported recall, extended periods of detailed observation, and in written diaries (Robson 2004;Reynolds 1991).
In recall interviews and diary-writing, children are asked to recount their activities in as much detail as possible, paying attention to the timing and duration of activities. However, both methods tend to under-report work because children forget tasks that they do not find important, tasks they are not allowed to do or find embarrassing, and tasks they do simultaneously with other work, such as childcare (Dyson 2014;Johnson et al. 1995;Robson 2004: 199). The recording of time use needs planning vis-à-vis the agricultural calendar, school holidays and even within a day (Robson 2004;Tudge and Hogan 2005).
Photography has also been used to observe children's day-to-day activities, including their work. For example, Bolton, Pole and Mizen (2001) conducted research into the working and economic lives of 11-16-year-olds, who were tasked with 'making photographs' 2 of their part-time jobs. PhotoVoice has become increasingly popular in the past decade to undertake research with children on a wide range of topics, including in the global South. In South Africa, for example, the method was used with children to understand their concept of 'self' (Benninger and Savahl 2016) and perceptions of the natural spaces around them (Adams, Savahl and Fattore 2017). However, the method goes beyond mere observation; it helps in 'making the familiar strange' to both researchers and participants and thus serves as a useful mediation tool to broaden discussions with participants, 'complementing, augmenting, confirming and enlarging insight from other methods' (Bolton et al. 2001: 517;Mizen and Ofosu-Kusi 2010). The method can also be adapted so that it can be used in participatory research with disabled children, such as in Sri Lanka and India (Wickenden and Elphick 2016).

Participatory and creative methods
Creative methods are often tools to engage children and other research participants to strategically democratise the research process by encouraging participants to become collectors of evidence and to encourage free expression (de Benítez 2011; Johnson, Hart and Colwell 2014; Mizen and Ofosu-Kusi 2010). These methods are considered more inclusive than other data collection techniques, in part because they involve play. They also often reveal 'subjugated knowledges' that participants would not articulate in aural or text-based media. Because images are multivalent, it is important that researchers do not make assumptions about their content, and that their creators are asked to explain and clarify their meaning(s) (Atkinson 2006;Ennew 2003).
Visual methods, including drawing, mapping and photography, are accessible methodological devices to use with children of all ages. Free drawing, open mapping and photo elicitation have been used with children and youth to gain a better understanding of place, space and young everyday lives (Bolzman, Bernardi and LeGoff 2017;Bowles 2017;Johnson 2011;Mitchell 2006). All three methods can be used in initial fieldwork stages to understand not only the context, but also central concepts of children's work. Later, more structured visual methods, such as 24-hour clocks of daily activities, body maps and Venn diagrams, can be used to gain more in-depth, nuanced understandings of causal pathways, connections and relationships. They can also capture children's and young people's feelings about their everyday lives and the social elements encouraging them to work (Hastadewi 2009: 481;Johnson, West and Gosmann, forthcoming).
Ruth Leitch (2008: 39) notes that, despite the common use of drawing in small in-depth studies, there are also good examples of its use for largescale audits. A large-scale policy consultation process in Northern Ireland, for example, collected drawings from 1,100 children and young people aged 5-18 (Kilkelly et al. 2004). Other large-scale and qualitative research projects have used drawing within their qualitative methodology to complement quantitative data (Crivello, Camfield and Woodhead 2009;Crivello, Morrow and Wilson 2013).
Performative methods and drama can involve dramatic play, individual or group mime, improvisation, or even a rehearsed performance. Katz (2004) asked children to make a model of their village on a patch of ground, then gave them a set of miniature toys (animals, machinery and people), asked them to identify each toy and then to show her 'life in the village'. All the children (10-year-olds) engaged in extended 'geodramatic play' using their models and the toys, during which Katz involved them in a running commentary that provided insights into their social and environmental knowledge (ibid: 283). Children often find it easier to communicate through drama than to answer direct questions. Often puppets or other role-play objects are used, creating a layer of distance and anonymity for the children (see Boyden and Ennew 1997;Johnson et al. 2014). This is particularly pertinent in research on CHW as it allows children to discuss forms of work that may be shameful in a more distanced manner. In South Africa, theatre-based research helped to unveil emotional challenges and notions of vulnerability among undocumented migrant youth in Cape Town (Opfermann 2020).
Written, workshop-based methods, including diaries, worksheets, activity tables and spider diagrams, are also methods to explore children's perspectives on their lives (Punch 2001a(Punch , 2001bThomson 2008). The success of these methods relies on a certain level of literacy, and on children having adequate time. Diaries, for example, can be more or less visual to include younger children or less literate children and young people, with increasing use of video diaries (Buchwald, Schantz-Laursen and Delmar 2009). Another option is to create a daily activity chart, or clock, in which activities are logged at different times, or recounted and drawn onto a timeline (Dachi and Garrett 2003). Worksheets can be prepared for the children to complete on different aspects of their lives, complementing drawings and photographs, or being complemented by interviews. Worksheets allow for more detailed information to be obtained on the issues which children have identified as important in their lives. Spider diagrams and activity tables are methods to capture the range of activities and work children do and the local geography of their work (Punch 2001a(Punch , 2001b. Focus group discussions (FGDs) represent a space for children to share their understandings and experiences in an interactive manner but without the pressure of having to engage with a researcher in a face-to-face interview (Gibson 2007;Hoban 2017). Participants in a focus group are often chosen because of their insights and views on different aspects of the research topic. For example, in a study of children's work in northern Ghana, Hashim organised FGDs with male and female senior secondary school students to discuss work, education, adulthood and childhood, and with adult women brewing pito (local beer) (Hashim 2004).
In Ethiopia, Abebe organised FGDs with adults to understand their views of children's work and childhood (Abebe 2008). Dyson's study involved several rounds of FGDs with key child informants and their friends in India, and two rounds with adults (parents of the key informants), focusing on children's work and on household budgeting (Dyson 2014).

In-depth interviews
In-depth interviews can help to explore a certain topic or issue in more detail. Life history or life cycle interviews, for example, aim 'to explore aspects of the social spaces of children and childhood' to understand the relationships that are central to children's psychosocial and material wellbeing (Abebe 2008: 57).
Participatory, creative and/ or ethnographic methods can inform the structure and nature of in-depth interviews. Using these methods does not merely constitute a process of piloting but also one of co-construction to iteratively build on ideas throughout the process of research. There are also many examples of interview processes that can be made more child-friendly and focused by, for example, carrying out interviews in peer pairs, or using interview props such as puppets, dolls and photos or pictures (Greene and Hill 2005;Johnson et al. 2014).
Semi-structured interviews focusing on children's everyday activities can be an abbreviated form of time-use allocation studies. This was the case in Abebe's study focusing on the activities that children had done, when, where and with whom, while also exploring gender and age differences and contextualising children's activities within the livelihoods of their families (Abebe 2008). Katz (2004) employed ethno-semantic interviews with 10-year-olds to elicit taxonomies of shared knowledge and an understanding of relationships and processes within their community. In these interviews, Katz probed children's practices. If a child mentioned that he/she had picked fruit in the course of a conversation about their daily work, she would seek more information about both the category of 'fruit' and of 'other things that are picked'. Through these interviews, she was able to establish a taxonomy of plant knowledge and place knowledge that helped illuminate the children's understanding of environmental processes and interrelationships (Katz 2004: 282).
Involving children in doing interviews may also work to break down the boundaries between the researcher and the researched. Hecht (1998) conducted what he called 'radio workshops' with street boys in Recife, Brazil. He gave them a tape recorder and microphone and asked them to interview each other. He found out that children often responded better to their peers, and they often asked better questions than the researcher, and questions that the researchers had not thought of (Boyden and Ennew 1997: 127). Chin (2007) found that the observations and discussions she had with children doing research with her produced more interesting and reflexive material than the interviews themselves. Children acting as researchers also entails risks that need to be carefully negotiated. A study that adopted 'participatory' docudrama with traditional Qur'anic students (almajirai) in Kano, northern Nigeria, found that the research afforded children the opportunity to voice their concerns and challenge stereotypes but also led to suspicion and accusations from within the community (Hoechner 2015).

Certification methods
Finally, we explore methods related to certification systems in agricultural value chains. These are mostly used outside of research settings but are employed by the private sector. In considering certification methods, we refer to tools that gather information about children's engagement with work within certification programmes.
Certification programmes emerged in the 1980s in response to consumer demands for sustainability and fairness and their willingness to pay for sustainably produced food items.
The first certification programmes concerned organic production, especially in Organisation for Economic Co-operation and Development (OECD) countries. Later, in the 1990s, Fairtrade emerged in response to a greater focus on fairness in value chain relations between smallholder producers in developing countries. At the same time, the retail sector in Europe started with certification schemes around food safety and good agricultural practices (GAP), which resulted in EurepGAP and later GlobalGAP standards. Certification and voluntary standard schemes are centred on tropical export crops, especially bananas, cocoa, coffee, sugar and palm oil. A significant part of the total production of cocoa produced in Ghana and Côte d'Ivoire is under one or more certification schemes (ISEAL 2019).
Four types of mechanisms for data collection can be used to glean insights into children's engagement with work, and these are discussed below.

Audit reports
Audit reports represent the main tool for information gathering within certification schemes and voluntary standards systems, such as Fairtrade and the Rainforest Alliance. Typically, control points in the audit differ for certification of individual producers (e.g. plantations or larger producers), and for group certification (e.g. where the production is scattered among many smallholder producers).

Group certification requires an accredited Internal
Control System (ICS), within which data on quality are managed by each group or firm. Medium or larger producers are audited directly, without an ICS. The other process is the Chain of Custody Certification, with requirements about how the product is processed and combined in value-added products, moving downstream from producers to consumers. For example, the Rainforest Alliance is currently developing a framework to enable routine audit processes to collect robust evidence without significantly increasing costs or administration for farmers, producer groups or companies. It is based on information that should already be available to auditors, such as field observations, maps, farm or group records, and interviews.
The quality of the audits (third-party certification) is a concern. Often there is a layered system that controls the accredited audit firms, who control the compliance of certification holders (especially producers). For example, the Forest Stewardship Council (FSC) has an agency, Assurance Services International (ASI), which provides this control-oncontrol.

Common core indicators
Despite the diversity of data collection across schemes, there is a tendency to harmonise information collected in certification schemes using common core indicators. ISEAL supported the development of linked, geographically referenced data sets for basic data collected in each scheme. The ISEAL common core indicators can be mapped against the indicators for the United Nations SDGs. 3 Some indicators directly refer to children's activities, including school attendance, distance to primary school, number of farms restricting the use of chemicals by pregnant women and children, food security (e.g. months and days of inadequate access to food), perceived change in quality of life, and perception of change in level of control over household decisions.
ISEAL works on a range of innovation projects to harmonise data flows within and between certification schemes (ISEAL 2019). This is partly to identify common and easily collectable data 3 See the full list and SDG mapping.
by implementers, auditors and evaluators, and to generate systems to store, link and analyse this information, and open its access to researchers. In addition, ISEAL has developed guidance for structuring data-sharing agreements for personal and sensitive data.

Outcome and impact evaluations
In addition to data generated within certification schemes by producers and auditors, the minimum requirement of ISEAL members is for certification schemes (scheme owners) to undertake at least one in-depth impact evaluation per year that addresses at least two questions: • Is the intervention producing the desired and intended sustainability outcomes or impacts?
• What unintended effects (positive or negative) resulted from the intervention?
In this case, the intervention refers to implementation of certification schemes or voluntary standards systems. Data on their intended and unintended outcomes constitutes a potentially useful source of information in relation to children's work. Data either covers all certification holders or a sample of certification holders, and a small number of studies are in-depth impact evaluations. For example, the Rainforest Alliance's approach to assessing its certification system (which was developed together with the Sustainable Agriculture Network -SAN) includes programme-wide monitoring, sampled monitoring and focused research. While data for the first two types of assessments are collected within operations and as part of audits, data for focused research tend to be collected by a third and independent party (ISEAL Alliance 2017).
In the past 15 years, these requirements have resulted in a large (perhaps disproportional) body of research on the impact of certification systems, including various systematic reviews (Blackman and Rivera 2010;Blackmore et al. 2012;Oya, Schaefer and Skalidou 2018;Schleifer and Sun 2020). Most of these studies focus on intended outcomes, like income and yield. Only a few discuss the impact or outcomes related to (children's) work as (intended or unintended) effects of certification.

Child Labour Monitoring and Remediation System
Another relevant data source follows from the Child Labour Monitoring and Remediation System (CLMRS), a relatively novel component of some certification schemes. A CLMRS tends to use local facilitators to collect in-depth information on all households in a region. Nestlé, Mars and other processing brands, for example, implement CLMRS as part of their voluntary standards systems. Nestlé (2019) reports that, by the end of 2019, it had identified more than 20,000 cases of child labour. This includes hazardous children's work, according to the definition used in Côte d'Ivoire, such as work with sharp tools and exposure to agro-chemicals. In other words, these systems offer purposive quantitative information on child labour or hazardous work within the production and processing of specific products.
3 Investigating prevalence, drivers and dynamics, and impact After having discussed the range of methods in detail, we move on to discuss their use in relation to investigating three key questions concerning CHW, specifically: (1) prevalence; (2) drivers and dynamics; and (3) impact. For each of these questions, we explore the opportunities or challenges presented by each method.

Prevalence
The question of prevalence refers to gaining insights into the scale and scope of different forms of CHW. Table 2 presents an overview of the opportunities and challenges of the different methods.
Surveys have been widely used to gain insights into whether or not children participate in work, and to generate numbers about prevalence at a wider (national or sub-national) scale. The ability to collect information across a representative sample allows for quantification of the occurrence of children's engagement with work and understanding the scale of the problem across age, gender and other lines of disaggregation. Indeed, household surveys such as LSMS, MICS, LFS and others represent key instruments for generating estimates about child labour and to monitor progress towards SDG 8 (UNICEF and ILO 2019). Quantitative components within mixed-methods designs may also serve to provide insights into prevalence of engagement with certain activities. Mixed-methods studies that include quantitative data from large-scale household surveys usually include prevalence estimates.
However, surveys are relatively ill-equipped to provide more nuanced understandings of children's engagement with work, and particularly CHW. We identify three reasons for this.
First, the rigid nature of survey questionnaires generally limits opportunities for understanding CHW. As noted by Bhalotra and Tzannatos (2003) and based on our review of survey methods, questions regarding categories of work tend to be crude and generally only allow for distinguishing between work for wages, work on family farms or enterprises, or domestic work. Surveys that underpin impact evaluations of social protection programmes also vary in the level of detail and the type of data that is collected about children's work (de Hoop and Rosati 2014). Purposive child labour surveys tend to be less bounded by stipulations within the ICLS resolution and therefore offer more flexibility. A downside of most of these purposive surveys -in terms of estimating prevalence -is that they are not nationally representative and so will only provide a partial picture.
Second, a prerequisite for identifying whether or not children engage in certain types of activities is their inclusion in data collection exercises. National household surveys are notorious for excluding the most marginalised groups, including children living on the streets or in institutions, and refugee populations (Bhalotra and Tzannatos 2003;Global Coalition to End Child Poverty 2019). This is particularly problematic when studying CHW as these children tend to be at greater risk (Bhalotra and Tzannatos 2003) but their work may be hidden from view.
Third, information is often provided by a proxy respondent rather than by children themselves, with caregivers answering on children's behalf. This may lead to inaccurate information: while caregivers may be well informed about their children's engagement in work, they may not have precise information about how children allocate their time or about working conditions; social and cultural values may also lead to under-reporting (Dammert et al. 2018). Equally, children may overestimate time spent on certain work activities or domestic chores (Dziadula and Guzmán 2020). While self-reporting is generally seen as more accurate and therefore more preferable (Desiere and Costa 2019), administering questionnaires to both adults and children will generate the most accurate results (Dziadula and Guzmán 2020).
Qualitative and participatory methods are vital for obtaining detailed and context-specific data about children's activities, their engagement with different forms of work, and the extent to which these are considered harmful and by whom. Prevalence mapping can be undertaken to gain a participatory understanding of how widespread certain forms of harmful child labour are. Such mapping can be built up and understood as an iterative process as more harmful work is made visible and trust is established with participants.
Data obtained through qualitative and participatory methods can also serve to develop survey questionnaires in order to improve their ability to gain insights into the prevalence of forms of CHW. Participatory and observation methods can help to develop categories of activities and time intervals that could be adopted in time-use surveys, for example. Tudge and Hogan (2005) describe an ecological approach to recording observations of children's lives with different tasks and activities being categorised and then recorded at defined time intervals by researchers who record what the child was doing. An example of a sequential qualitative-quantitative study to improve prevalence estimates of forced labour is the ILO's International Programme on the Elimination of Child Labour (IPEC) and Special Action Programme to Combat Forced Labour (SAP-FL) (ILO/ SAP-FL/ IPEC 2012).
This study started with qualitative investigations and later developed and trialled quantitative survey tools.
Finally, certification methods can also provide insights into children's participation in certain types of activities. These certification systems may offer information beyond standard categories (adapted to local context and needs). The CLMRS, for example, collects metrics about the extent to which children work in agriculture and how many are involved in hazardous tasks (based on the CLMRS's own definitions). This offers information about prevalence within a certain industry or value chain. However, reliability of data may be a concern when using this information.

Drivers and dynamics
Different methods offer different strengths and challenges for understanding drivers and dynamics of CHW, as presented in Table 3.
Surveys are widely used for studying drivers and dynamics of child labour and children's engagement with work. Macro-level studies focus on correlates at country level and are mostly premised on cross-country data. The Understanding Child Work (UCW) programme, for example, considered country-level variables such as gross domestic product (GDP) per capita, ratification of ILO Convention No. 138, exports of clothing and textiles, and the Fragile States Index to understand differences in trends across countries (UCW 2017).
Micro-level studies are much more common and typically explore the role of demographic and socioeconomic characteristics of households and their members in explaining patterns of children's engagement with work. In Bangladesh, for example, the Household Expenditure Survey was used to investigate the role of household poverty and wealth in child labour. Regression modelling was used to estimate associations between independent variables such as household income and educational achievement of households and the dependent variable of children's work (Amin, Quayes and Rives 2004). The Young Lives study has led to research on the determinants of work participation and school attendance and their trade-off in Ethiopia (Haile and Haile 2012). Schoolbased surveys have also been used to understand how children's engagement with work is associated with academic performance (Guarcello et al. 2005).
The caveats outlined in section 3.1 in terms of the ability of surveys to capture the prevalence of children's hazardous or harmful work also hold for drivers and dynamics. The sets of questions that are included in surveys are often too limited to allow for detailed understanding of factors that are associated with, or cause, CHW. It is also important to note that due to the cross-sectional nature of many surveys, most studies allow for investigating association but do not offer insights into causality. Exceptions include studies that use longitudinal data and econometric methods that allow for estimating causal effect. In Ghana, for example, three waves of the Ghana Living Standards Survey were used to investigate determinants of child labour (Blunch, Canagarajah and Goyal 2002).
Qualitative and participatory methods -particularly when used in combination -can uncover 'subjugated knowledges' and everyday granular realities and constraints that are not necessarily articulated in surveys but are vital to understand why children engage in work. For example, photography and creative methods are suited to making different aspects of children's lives and their feelings more visible. Drama could be used to animate discussions about work, unequal power relations, expectations of labouring, and children's ability to influence their workday and load. FGDs, interviews, participant observation, diaries and mapping exercises all present vital tools for unveiling underlying choices and constraints in terms of work, both from the perspectives of children and others.
As is the case for survey methods, longitudinal data would allow for greater uncovering and understanding of factors playing into children's engagement with work. At present, there are no longitudinal mixed-methods studies specifically on the dynamics of child labour (e.g. how children's workloads change over time, or how changes in a household's poverty level may affect children's labour participation). This shortcoming has also been highlighted by other authors (Camfield 2014;Ibrahim et al. 2019;Kuimi et al. 2018).
Relatedly, narratives of change are necessary to get insights into the drivers and dynamics of children's work. A CLMRS may provide part of these narrative accounts within households that are at risk. Local facilitators are well suited to identify illustrative cases, for example, particularly as children are not attending school or may not be registered in the health post when injured or ill. In several of these Requires careful sampling to ensure a range of perspectives across respondents; requires time to build capacity in skills and ongoing ethical procedures to facilitate some of these methods

Localised longitudinal information can support analysis of changes in household conditions and harmful work
Local facilitators in CLMRS can help collect more in-depth information but are unskilled as researchers Source: Authors' own.
CLMRSs, a social worker, paid by the company, is responsible for visiting households. They raise awareness about the tasks that children of that age are considered to be capable of doing or not doing. However, a downside of working with local facilitators is that they are relatively unskilled as researchers.

Impact
We explore how different methods can shape an understanding of how CHW impacts various aspects of children's lives, and how interventions impact CHW (see Table 4).

Impact of child labour on children's lives
Survey methods are commonly used to assess the impact of child labour or children's engagement with work on different aspects of their lives.
Many studies are particularly interested in associations between work and education. For example, NCLS data from 12 countries was used to investigate associations between child labour and educational attainment (ILO 2015). Young Lives data underpinned a study of the impact of child labour on educational attainment in Vietnam (Mavrokonstantis 2011). Several mixed-methods studies (Orkin 2012;Woldehanna, Jones and Tefera 2008) also explored the impact of child labour on school attendance in Ethiopia. Qualitative and participatory studies can uncover intended and unintended consequences of work, placing these within contextual understandings of harm.
Four observations are important. First, as noted above, most analyses are based on cross-sectional data and only allow for gaining insights into associations but not causality. Fourth, the issue of temporality is key in understanding how work affects children, and whether or not it may be harmful (Maconachie et al. 2020). Work may only cause harm if children are exposed to a certain risk associated with that work over a longer period of time, and harm may also present itself long after children have stopped engaging in this work. For example, agro-chemicals may only cause harm if children are exposed to them over a longer period of time, but its effects could be immediate (e.g. chemical burns), medium term (e.g. respiratory problems) and/or long term (e.g. affecting reproductive health). While the range of methods are relatively well-equipped to pick up intensity of exposure through studying time use, few methods have enough of a longitudinal perspective to pick up on medium-to long-term effects, particularly if the potential for those effects is not yet known.

Impact of programmes on child labour
In many impact evaluations, surveys are central to the research design and constitute the primary data source for estimating programme effects, particularly in (quasi-)experimental settings. Evaluations cover programmes that have the reduction of child labour as a primary objective (e.g. educational interventions) and programmes that have it as a secondary objective (e.g. social protection).
A notable observation in relation to quantitative impact evaluations is that child labour (or children's work) tends to be loosely defined. Studies -and their underlying surveys -are often designed without clear reference to international guidelines or academic literature that problematises dominant understandings of child labour or children's engagement in work. This is certainly the case in relation to social protection. Evaluations of social protection programmes and their effects on child labour rarely follow the ICLS resolution (Dammert et al. 2018). Notions such as child labour or children engaged in productive activities are used interchangeably, with some evaluations denoting any type of work as child labour (ibid.). Even evaluations of programmes that focus squarely on reductions in child labour concede that there is no agreed definition of child labour and adopt their own operationalisation, such as in relation to educational programmes in Panama (Andisha et al. 2014).
Qualitative and participatory methods are crucial for uncovering intended and especially unintended impacts of interventions. Impact assessments that are child-centred can use a range of already tried and tested visual, narrative and mobile methods, as already discussed, including photo narrative workshops, sorting and ranking different work activities, observation of children's and young people's work, and using matrices with children's indicators for health and wellbeing. Time allocation methods will inform the understanding of how work fits with children's everyday realities, and genderdisaggregated data and analysis will be important. These methods combine observation and statistics. Impact work can be carried out with children, parents and other stakeholders in employment and the informal sector to compare different intergenerational perspectives. Drama could also be used to elicit reactions to interventions aiming to abolish child labour.
A CLMRS that exists within certain certification systems may offer specific information or scope for collecting data about the impact of certification on children's engagement with hazardous or harmful work. These multi-stakeholder and areabased approaches are one of the more promising interventions to address the issue of hazardous child labour (ICI 2011;ILO 2018). Remediation activities are at the heart of the efforts of CLMRS and involve supporting children, their families and communities to remove children from a situation of risk. The purpose is twofold: to try and prevent children from doing hazardous work in the first place; and to help children who are engaged in hazardous work to stop. The majority of remediation activities to date have focused on education, activities to improve family income, and assistance with farm-related work (Nestlé 2019). With the CLMRS still in its piloting phase, implementors of these initiatives are looking for research that can help them to perform these activities more effectively and in such a way that the costs are covered as part of the business strategies and sharing of risks and rewards in the supply chains. Difficult to attribute impact (although can be useful to assess contribution); questions can be scaled (to be administered to large samples) but methods need to be combined with other methods to gain full insight

Provide insights into impact of certification systems on CHW
Data may not be reliable Source: Authors' own.
4 Mixed-methods design 4 There are many variants of each mixed-methods designs, for further details please see Creswell et al. 2003. 5 About Verité.
Next, we consider the use of mixed-methods design in studies of child labour and children's work.
For the purposes of this review, we define the term 'mixed methods' as the combination of quantitative and qualitative approaches (and excluded studies that only combined multiple qualitative or quantitative approaches). Also, for feasibility purposes, we considered only studies that focused on child labour or children's engagement with work as a main outcome of interest. A total of 10 studies were identified to fit these criteria.
In this section, we provide an overview of mixedmethods designs that were used to underpin the studies identified, and assess the opportunities and challenges for using mixed methods when studying CHW.

Mixed-methods studies of child labour
We follow Creswell et al.'s approach to mixed methods study designs for the reviewed studies, highlighting four major types of mixed-methods designs (Creswell, Plano Clark and Garrett 2003). 4 First, in the triangulation design, both qualitative and quantitative data collection and analysis take place concurrently but separately. The findings from each method are then brought together in the final interpretation phase. Second and third, the explanatory and exploratory designs constitute sequential two-phase mixed-methods designs whereby qualitative and quantitative data collection and analysis take place at different times and the findings explicitly aim to build on each other. For example, in an explanatory sequential design, the qualitative data collection and analysis help to explain the initial quantitative results. Fourth, the embedded design is a mixed-methods design in which one data set provides a supportive, secondary role in a study based primarily on the other data type.
Two of the identified studies that used a mixedmethods design to study child labour as one of their main outcomes of interest (Ghorpade 2017;Zakar et al. 2015) employed concurrent triangulation design, whereby quantitative and qualitative data collection took place at the same time and findings were combined in the final analysis phase.
Half of the studies used a sequential design, mostly starting with the analysis of quantitative survey data and followed by in-depth qualitative work for more nuanced insights. Two studies used sequential mixed-methods designs with multiple qualitative and quantitative rounds that sequentially informed and built on each other (Orkin 2012;Verité 2016). The latter study presented a major multi-country research project on forced labour conducted for an international civil society organisation (called Verité 5 ). The project used a flexible approach with a variety of qualitative and quantitative methods. Country teams could select and adapt methods depending on their on-theground research needs and challenges. The authors of this study emphasised that the mixed-methods approach provided them with much-needed flexibility and the ability to adapt data collection efforts to dynamic and often insecure contexts.
Another study (Al Ganideh and Good 2015) used a sequential explanatory design for the data collection but employed a more pragmatic, iterative approach during the data analysis and triangulation. In practice, this meant that the researchers followed up hypotheses that emerged during the analysis of one data source with the other data source and vice versa. This more flexible approach helped the authors to provide more comprehensive insights and explanations than a strict one-way sequential approach might have done. A well-documented drawback of this type of mixing methods is the length of time necessary to develop and adapt methods in a sequential manner (Creswell et al. 2003) One study (Bhatia et al. 2020) used an embedded design with the quantitative approach being the primary data source. A small number of qualitative stakeholder interviews were undertaken merely to expand the quantitative findings. Another study used an embedded design with a primary qualitative component and secondary quantitative survey (O'Kane, Barros and Meslaoui 2018).
Only one study used a more innovative approach to mixed-methods research (Kiss et al. 2020), using a realist evaluation design with Bayesian network analysis to explore causal pathways to and drivers of forced labour in Nepal. The authors criticised traditional, linear mixed-methods approaches for simplifying the underlying causes of forced labour and failing to acknowledge its interlinked complexities. As a consequence, these traditional studies often concluded with inaccurate results and had limited explanatory power (ibid.).

Opportunities and challenges of mixed-methods design
In this section, we assess the potential and challenges when using a mixed-methods research design for studying children's engagement with work, and particularly CHW (see Table 5). Overall, mixed-methods approaches can be powerful as they combine strengths of various methods. They often help to challenge perceptions and assumptions about children's work and thus can facilitate a more holistic understanding of children's engagement in harmful work. The review of existing mixed-methods studies of child work also shows that this potential has so far been largely underexplored. The level of interwovenness between the quantitative and qualitative components in the retrieved studies was generally weak. In the majority of mixed-methods studies, the quantitative and qualitative components were conducted separately and, to a large degree, independently of each other.
With respect to prevalence, mixed-methods design offers real potential for making estimates of child work more meaningful and reliable. As noted in one study, NGO members stressed that national-level prevalence data were important to highlight the magnitude of child labour for their advocacy work but that they had limited use with regards to gaining fine-tuned insights to guide action and programmes. Local-level data and qualitative approaches were perceived as necessary for this (Bhatia et al. 2020). It follows that mixed methods offer promising opportunities for estimating prevalence of CHW by first gaining more detailed insights into working conditions and then estimating prevalence using quantitative data.
As noted in section 4.1, various mixed-methods studies have used both qualitative and quantitative methods to gain insights into the conditions that children work in (Bhatia et al. 2020;Al Ganideh and Good 2015). Nevertheless, few studies have made full use of the opportunity to preface survey data collection with in-depth qualitative data generation to map the prevalence of children's engagement with work from a more nuanced perspective.
In terms of the drivers and dynamics of child labour, many purely quantitative studies neglect the heterogeneity of child labour, which can significantly reduce the usefulness of findings to inform policy and practice (Krauss 2017). Mixedmethods approaches can facilitate the identification of meaningful sub-groups of child workers and what influences their participation in work, thereby ensuring that research is more inclusive. In Ethiopia, Orkin (2012) employed a sequential, multi-phased mixed-methods design to explore the drivers of both child labour participation and school attendance. Qualitative fieldwork with parents and children was used to identify characteristics of work and school that influenced participation, which was then used to inform and improve analysis of quantitative models on intra-household bargaining with regards to children's time allocation to either school or labour. In other studies with sequential design, the quantitative analysis proposed one or more potential drivers for child labour while the qualitative data was then able to provide details on the potential causal mechanisms behind the observed association (Shaffer 2013). For example, based on the econometric analysis, Woldehanna et al. (2008) found that children with highly educated mothers were more likely to work. Qualitative findings indicated that educated mothers were often more likely to work outside the home, thereby increasing domestic work for their children at home.
A considerable shortcoming, also observed in relation to other methods, is the lack of longitudinal data. This hampers the ability to explore what drives children's engagement with work over time, and limits the ability to understand the impact of children's work. Several authors noted the lack of long-term mixed-methods studies on the mediumand long-term consequences of work on children's health and wellbeing (Ibrahim et al. 2019;Kuimi et al. 2018).
The Young Lives study is a notable exception to this and has underpinned various investigations into the impact of children's work. Several studies (Orkin 2012;Woldehanna et al. 2008) explored the impact of child labour on school attendance. Drawing on both qualitative and quantitative evidence, the authors found that work and school attendance may be successfully combined depending on the time each activity takes and the characteristics of each activity. A potential pitfall when it comes to mixing methods is that tools may be premised on different understandings of what constitutes child labour or harmful forms of work, thereby potentially limiting the extent to which findings can be combined and complement each other. At the same time, these alternative views can facilitate a deeper understanding of why impacts do or do not play out.
Finally, an obvious but necessary observation from this review points to the overall lack of mixed-methods studies on children's engagement with work. This seems to echo the perennial and persistent divide between quantitative and qualitative research observed within development studies (Jones and Sumner 2009). Findings suggest that quantitative studies still mainly focus on assessing the prevalence, drivers and impact of child labour. By contrast, qualitative and participatory research seems more concerned with investigating children's experiences of labour and the dynamics and complexities surrounding it. We also find that the majority of studies focus on obtaining larger-scale data that can be contextualised with more qualitative methods. Relatively few studies adopt fully integrated designs or make use of child-centred and participatory methods in combination with quantitative methods.

Implications for ACHA
This review leads to reflections about implications for ACHA and its research design. Generally, the review of methods shows that there is real potential for ACHA to do something new, innovative and exciting from a methodological point of view. The review identifies two research gaps that ACHA can begin to fill. First, despite the wealth of research on child labour and children's work, few studies use a truly integrated mix of methods. This integration would enable them to think beyond and challenge standard notions of children's engagement with work. Second, only a relatively small body of literature (across all research looking at forms of child labour and children's engagement with work) seems to be concerned with children's hazardous and harmful work. This literature is primarily informed by smaller-scale ethnographic and participatory research due to the complexities and sensitivities surrounding those types of work. ACHA has an opportunity to adopt a research design that integrates methods across disciplinary divides in more holistic ways and, in doing so, to begin to understand the breadth and depth of children's harmful work in agriculture.

Research design principles
Following the review of methodological experiences, we frame implications for ACHA as research design principles, which would inform a more detailed research design. This would involve the following steps. • Giving space and weight to children's voices: Research that actively engages children and recognises their expert knowledge produces better data and identifies gaps in needs and priorities -and often produces unexpected findings (Johnson and Lewin forthcoming;Mizen and Ofosu-Kusi 2007;Van Blerk et al. 2009 (Maconachie et al. 2020). In Ghana, the research focuses on three relatively distinct supply chains (cocoa -international; inland fish -entirely national; shallots -entirely national). Research design needs to be adequately contextualised -for example, in terms of which stakeholders will be included as research participants. Contextualisation of research also requires taking account of differences across spaces and places Source: Authors' own.

Certification methods
Qualitative/ Participatory Action Research (PAR) methods

Drivers and dynamics
Qualitative/PAR methods

Impact
Qualitative/PAR methods (e.g. studying the forms of CHW in small-scale fisheries around Lake Volta in Ghana may also need to study the living conditions in areas from which children move to Lake Volta to engage in work).

CHILD-FOCUSED, INCLUSIVE, ETHICAL
• Adhering to ethical protocol and principles: Research design and individual methods need to be fully in line with ethical protocols, procedures and practice (Johnson 2020). Such ethical protocols are to be developed through local dialogue in keeping with the principle of building on capacity of local researchers and participants.

Methodological landscape
In line with the research design principles, we propose a rough methodological landscape that offers parameters within which ACHA's research design can be developed (Figure 1). The proposed methodological landscape differs from the existing predominant use of methods in researching children's engagement with work in a few ways: • The mixed-methods approach is more holistic and all-encompassing, fully integrating survey methods, qualitative and participatory methods and certification methods.
• Relatedly, this landscape gives greater weight to qualitative and participatory methods.
The complexities and sensitivities involved in researching CHW merit the use of such methods, particularly in the early stages of the research and to establish prevalence.
• Stronger linkages are in place between methods, aiming towards an integrated mixed-methods design as opposed to purely sequential or parallel designs. The bi-directional arrows propose an iterative research process whereby, for example, data from qualitative and participatory methods feed into survey design and findings from survey data can feed into ongoing ethnographic activities. As already noted, the exact combination of methods may be adapted over the course of the programme.
• Methods are integrated across the research process to make full use of the learning from individual methods and the expertise of respective researchers from design through to uptake of research findings. Crucially, this requires ample allocation of time in order to make full use of learning opportunities created through the research.