Qualitative secondary data analysis: Ethics,epistemology and context

I Introduction

As part of a growing ethic of the scientific value of data sharing, and related to remarkable advances in electronic infrastructure for archiving and sharing of such data, there has been a drive towards promoting and facilitating the reuse and secondary analysis of qualitative data (Bishop, 2009; Bishop and Neale, 2012; Van den Eynden et al., 2011). The UK National Data Strategy (2009–12) has contributed to practices of data sharing and an infrastructure for doing so, and funders within the UK and across different country contexts increasingly require researchers to consider data sharing as part of their research designs and funding proposals (Bishop, 2009; Bishop and Neale, 2012; Neale and Bishop, 2010). In the UK, the central repository for social science and humanities research data is the UK Data Archive. Qualitative researchers deposit anonymized data which meet standardized archiving requirements relating to its format and ethical consent, for example, and they may themselves require certain conditions be met in allowing access to their data. There has been a significant growth in possibilities for reuse of qualitative data, although arguably such possibilities remain underutilized. There is minimal evidence of qualitative secondary analysis in the development literature and interesting questions present themselves regarding the scope and value of such an approach within development research.

Qualitative secondary analysis entails the use of already produced data to develop new social scientific and/or methodological understandings. There are a variety of reasons why researchers may undertake qualitative secondary analysis. They might return to their own data to address new questions which were not a focus of the original study (Gladstone et al., 2007); they might seek to relate their own primary research or data to existing data resources; they might ask new questions of extant qualitative datasets which they had no role in producing, or bring different datasets into conversation with one another (for example, Bishop, 2007); they might seek to develop insight into hard-to-reach populations or sensitive topics (for example, Fielding and Fielding, 2000); or they might undertake a critical analysis of the embeddedness of methodology and explanation in historical and theoretical context (for example, Bornat, 2010; Savage, 2005; and Walkerdine and Lucey, 1989). Some commentators ask whether the practice of secondary analysis could contribute to understanding the generalizability of findings from qualitative studies, potentially enhancing the scope for findings from qualitative research to be cumulative (for example, Fielding, 2004).

In this article, I draw on ESRC Timescapes, a 5-year long programme of work centring on seven primary qualitative longitudinal research projects, run by project teams in five different universities. ¹ The projects were independently conceived, and some were in place before Timescapes commenced. They were brought together and cohered due to a range of shared substantive interests, in biography, life course and life course transitions, familial relationships and intergenerational dynamics. They were all qualitative longitudinal projects which engaged with social processes as temporal and dynamic, and utilized a range of methods oriented to exploring time in different dimensions, including experiential, biographical and historical. Part of the aim of Timescapes was to develop resources to facilitate qualitative longitudinal data reuse. The ESRC Timescapes Archive of qualitative data (a satellite of the UK Data Archive) comprises data from the core projects and affiliate project data. Timescapes, additionally, itself undertook a range of secondary analysis activities involving joint work across project teams, a programme of training and a dedicated Secondary Analysis Project run by the current author and a research fellow, across a two year period, within the final stages of the overall programme. It is work from this project which is discussed here. ² I commence with a brief consideration of issues of ethics and epistemology in the conduct of qualitative secondary analysis. In Section III, I address some practical and conceptual issues, particularly as these pertain to how we situate data and to its contextual embeddedness. Such questions hold particularly acute ramifications within development research. In Sections IV and V, I describe how the Secondary Analysis Project team addressed these issues in two substantive research examples. Timescapes was a UK project and does not speak directly to international development issues. I hope to illuminate some methodological and linked conceptual questions, and encourage critical reflection on the scope for undertaking secondary analysis within development research.

II Qualitative data reuse: Resources, ethics and practices

The practice of qualitative secondary analysis has been accompanied by some controversy, partly relating to competing ideas about ethical practice. Qualitative research typically entails very different relationships between researchers and participants than do survey-based strategies, and such relationships are inflected in the data. The relationships in which qualitative research is itself embedded, the ways in which flexibility rather than fixity features in data-generation strategies and the embodied interpretations of interaction and meaning generate particular challenges for analysts. Data are generated through interactions between researchers and participants and are shaped in ways which relate to the project design, researchers’ disciplinary assumptions, theoretical inclinations and methodological decisions. The contextual embeddedness of data engenders ethical and epistemological challenges to analysts, which I discuss in turn.

Critics of qualitative secondary analysis have foregrounded ethical risks for the participants (for example, Mauthner et al., 1998; and as reported by Broom et al., 2009). Data are generated through personal interactions with participants, often involving trust and a duty of care towards participants’ data on the part of the researcher. This has led to concerns regarding the use of such data by others and shaped debate about ethics and informed consent. Where future use of data is unknown, informed consent can be secured in relation only to generic statements about the uses to which such data might be put (for example, that the data will be available for research and teaching purposes). Another area of concern has been that researchers themselves are heavily implicated within qualitative datasets, yet their own rights may be under-regarded (for example, Mauthner and Doucet, 2008). Whilst these sorts of concerns reflect real challenges they are challenges which, many argue, can be addressed effectively (Bishop, 2009; Kuula, 2010; Mason, 2007). Bishop (2009) and Bishop and Neale (2010) argue that the debates about ethics in qualitative secondary analysis should be framed more broadly than hitherto such that alongside rights (to confidentiality, anonymity, integrity in the uses to which data are put) there should be greater consideration given to duties, with reference to different stakeholders, including research participants, primary and secondary researchers, project partners such as service providers and the public, albeit that their interests will be weighted very differently (Bishop, 2009; Neale and Bishop, 2012; Neale and Hanna, 2012).

Alongside, and partly overlapping with, ethical issues are epistemological ones relating to the possibility of effective interpretation and analysis of data by those who come to it from a distance, uninvolved in the process of data creation. Critics maintain that the practice of reuse and of secondary analysis occurs on thin foundations, without the immersion in data, meaning and context that qualitative researchers deem an essential part of grounding knowledge claims (for example, Mauthner et al., 1998; Mauthner and Parry, 2010; also cf. Hammersley, 1997; Mauthner and Doucet, 2008). A particular concern relates to a notion that archived data might be treated as foundational, neutral and effectively ‘cleansed’ of the contextual, conceptual and interactional contexts in which they were produced and through which they should be understood (Mauthner and Doucet, 2008). In this view, presence at the point of data creation and knowledge of the proximate contexts in which it occurs are deemed crucial to an authentic understanding and analysis of data.

Before arguing the case for secondary analysis, it is important to stress that, where the aim is to derive further substantive meaning from the data generated by the primary researcher, diverse kinds of project and data may or may not lend themselves to secondary analysis. In ethnographic inquiries and fieldwork in which the researcher is immersed in the study setting and evolving theoretical understanding as part of the logic of enquiry, data are very much the product and possession of the researcher. There will be narrow limits on the possibilities for reanalysis by third parties. In contrast, a series of semi-structured interviews generate data which are more independent of the researcher. Here, the primary researcher’s own role and assumptions are more visible to the data reuser, and the process of data generation is more transparent. Here, secondary analysis is potentially productive. However, with respect to even relatively structured qualitative enquiry, the researchers’ presence and aspects of proximate context permeate the data in ways not always immediately (or sometimes possibly ever) evident to subsequent users of the data. That is, aspects of context will remain opaque or even invisible to secondary analysts and may undermine the validity of their interpretations. Many have countered that whilst secondary users lack immediate knowledge of the research settings, there is no prima facie reason why the primary researcher has a uniquely privileged awareness of the situatedness of the research endeavour (for example, Bishop, 2009; Hammersley, 2010; and Irwin and Winterton, 2012). It may be that distance itself sheds analytic or critical light. Many concur that secondary analysis is a productive undertaking and potentially (as with extant qualitative datasets) underutilized. In our own secondary analyses, we worked with semi-structured interview data produced by others in their own projects, within culturally familiar settings. However, it is important to pause at this point and ask if particular questions and methodological difficulties arise for development researchers.

A central challenge for secondary analysts is that they are ‘at a remove’ from the data and the contexts in which they were generated. Where analysts are returning to their own data to ask new questions, this distance may work in their favour as they have both proximate knowledge of the data and a form of critical distance. Where analysts are studying other people’s data, they need particular strategies for seeking to understand aspects of context which are important. These issues have been widely debated and discussed in a Western context but, to my knowledge, not in relation to majority world research contexts. The concerns which have been debated may engender particular difficulties in development research. For example, in development studies, much qualitative research is anthropological, ethnographic and fieldwork based. Analysis by others may be irrelevant, and whilst researchers may return to their own data generated in earlier projects, in contexts of profound and rapid change, this too may risk redundancy. A potentially more troubling question is the one about knowledge, error and power relationships in development research and development projects (cf. Chambers, 2010). Engagement with local communities’ development needs has shown the value of solutions which are ‘ground up’ but also at least partly generated and morally owned by communities themselves (for example, Chambers, 2007 and Kar and Pasteur, 2005). Long-standing participatory action approaches to research in developing country contexts aim to build researchers’ understanding of culturally specific social arrangements, norms and values, and to empower participants through the process of undertaking research. It is hard to see what secondary analysis would add to such approaches, nor is data recorded and preserved in such a way that would make secondary analysis possible. However, in more conventional research contexts where data might stand as a ‘body of evidence’ amenable to secondary analysis, there remains a concern that distance from cultural and contextual specificity may magnify the problems of being ‘at a remove’ from the contexts in which primary data are generated. It may be that the only way of handling this is to be constantly critically reflexive and analytic, and use such data and analysis to generate questions. Potentially perhaps, such questions could then be taken back to their originating contexts, through dialogue with primary researchers or through new primary research.

In the following section, I briefly discuss some conceptual as well as practical issues relating to the conduct of secondary analysis. The issues discussed are not unique to secondary analysis, but they present particular challenges in working with secondary data. The first relates to how we understand data to be situated in respect of wider patterns and processes, and the second relates to issues of the embeddedness of data within diverse project contexts and how, then, we can bring diverse datasets into meaningful analytic conversation.

III Practical and conceptual issues in working with new datasets

Whilst qualitative research may often entail relatively small numbers of cases, so too, it typically entails very extensive amounts of data. Bearing in mind how secondary analysts need to familiarize themselves with data, I commence with a note on some practical issues which may also help inform primary researchers’ practices in making available their own data for effective reuse. Primary researchers need to budget costs and time for anonymizing and archiving transcripts and other data if necessary, and can helpfully supply comprehensive metadata (data about data) to facilitate reuse. Reusers need to orient themselves to the project through accessing available literature on the project by the originating researcher or team, and any overviews of data and resources archived by the originating researcher(s). They need to understand the research objectives, design and the research questions and methods used for data generation (including interview schedules or other data elicitation tools). They need a grasp of the sample, including knowledge of the sampling decisions and how they related to the research questions, whether the desired sample was achieved and how it related to a wider population and/or to theory. In addition, it is useful to understand any implicit as well as planned ways in which the sample was structured. For example, does the basis on which people opted into the study have implications for understanding the extent to which participants ‘stand for’ a wider population, or to which they provide insights into wider experiences? Primary researchers might consider documenting their insights into this. Further considerations relate to the amount of contextual information supplied by primary project researchers. This might relate to the data itself (for example, annotations within transcripts), to the fieldwork context ³ and to notes on wider social contexts (Bishop, 2006; Irwin and Winterton, 2011a, 2011b).

I turn now to conceptual issues in the conduct of secondary analysis and consider the grounds on which secondary analysts might seek to build knowledge claims. One issue here relates to conceptualizing and situating data specificity. In seeking to situate data, it is important to ask how it ‘stands for’ the processes under investigation. These processes are complex and multifaceted and points of data form small, and theory laden, clues in respect of such processes. We could accept that qualitative data are, in any event, not representative and ask if and why it matters which participants, or what subsample, we pick out for more detailed analysis (cf. Geiger et al., 2010). However, whether they are working with a comparative strategy or one based on developing theoretical cases, qualitative researchers commonly seek an understanding of how people are situated within study samples in order to grasp the contexts underpinning diversity, and hence insights into conditions and causes. Any selective analysis of cases needs a thorough understanding of how such cases relate to the dataset as a whole. Savage, for example, notes the risks of so-called ‘juicy quote syndrome’ (cited in Geiger et al., 2010) in which analysts place undue weight on eye-catching data. We might caution also against juicy case syndrome. There are, of course, ways in which singular cases may be enormously informative but we need a situated sense of what they reveal (Emmel and Hughes, 2009). A case may be deemed theoretically rich but we need grounds for ensuring we have grasped key features of its situatedness and context in properly adducing causal processes. A single case cannot ‘stand’ only because it beautifully exemplifies a particular theoretical or policy claim. The links which are drawn between experience, meaning and context will be more rigorous, and available to scrutiny, if we show how we have brought them into conversation or comparison with other cases. Analysts can thereby usefully illuminate the rationale behind their choice of cases for detailed analysis, or the choice of iconic cases in published work.

A different way of situating data would be in respect of wider patterns extant in the population. For example, we might situate data with reference to quantitative evidence derived from large samples. Timescapes existed as a qualitative longitudinal study which was wholly independent of the UK large datasets programme. We sought to design in some points of connection; so whilst the studies are independent, there might be scope for working across datasets in a meaningful way. An example of this is drawn from the Timescapes Young Lives and Times (YLT) study in which a project survey was conducted across the same education authority area as the qualitative research (Irwin, 2009). The questionnaire included questions which mapped on to national-level datasets, such as the Longitudinal Study of Young People in England (LSYPE). Participants in the YLT qualitative longitudinal study also completed the survey questionnaire. This means that we could see how, with reference to particular indices (including attitudinal data), the qualitative longitudinal study participants were situated with respect to population heterogeneity evidenced in the survey. The strategy meant we could orient to a more case-based, comparative analysis, where cases could be located in respect of aggregate patterns (for details, see Irwin, 2009).

Context has been construed in quite particular ways in debates about secondary analysis and, with Mandy Winterton, I have argued that it would be beneficial to supplement this with further attention to a set of middle-range issues pertaining to how research problems are conceptualized and framed (Irwin and Winterton, 2012). In order to assess data as evidence, we need to understand how it offers a lens on the phenomena under investigation, or which facets of our research questions are being addressed (Irwin, 2008; Mason, 2002). There are important questions about how salient contexts are accessed through data-generation strategies. The concern with grasping relevant contextual data lies near the heart of the recent expansion of interest in data-generation strategies which move away from established interview and participant observation traditions. Many seek more ‘ground up’ strategies, such as visual methods, walking interviews, relational mapping exercises and other techniques which seek to get closer to people’s lived experiences and ‘real lives’ than may be possible in conventional interview formats (for example, Emmel and Clark, 2009 and Mason and Davies, 2009). Many of the Timescapes projects were themselves exploring how the longitudinal perspective allows points of comparison through which contextual relevancies are thrown into relief (Emmel and Hughes, 2009; and cf. Holland and Thomson, 2009). There has been a growth in participatory methodologies in development research over recent decades (Chambers, 2010). These strategies exemplify methods for accessing a more valid understanding of social actors’ own experiences, perceptions and actions. Nevertheless, assumptions about salient context are embedded within all data-generation strategies. Research designs, methodological strategies and data collection tools access multifaceted problems in potentially quite particular ways. Particular conceptualizations of the nature of the social phenomena and of salient contextual information are embedded in methodological decisions, shaping the available data and encouraging particular ‘readings’ of the social phenomena being researched (cf. Chambers, 2007). When we bring data from diverse projects into comparison, we can see evidence of the impact of diverse project contexts on the data generated (for example, see Irwin et al., 2012). The embeddedness of data in project contexts does not mean we cannot work across them, but this is a conceptual task as much as an empirical one.

IV Developing a secondary analysis of qualitative longitudinal data: Tracking evolving expectations and links to social inequalities

In this and the subsequent section, I develop two examples which illustrate aspects of the conceptual points discussed earlier. In the first example, I briefly illustrate a qualitative longitudinal analysis drawing out some themes relating to how individual cases are situated in relation to wider evidence on pattern and process. One strategy for undertaking analysis of qualitative longitudinal data is to follow a case-based analysis (see Edwards and Irwin, 2010, for examples based on Timescapes projects). Researchers may move between a cross-sectional form of analysis, working across data from one wave of interviews, and a case-based analysis in which they work longitudinally. They may do these conjointly in order to analyze the patterning of biographical continuity and change; or they may work with longitudinal case studies as these provide analytic purchase on some set of wider conceptual or theoretical questions (for example, Holland and Thomson, 2009; Thomson 2007; and Thomson et al., 2002). In the example used here, I take a case-based approach, within an analysis which situates individual cases with reference to wider patterns within which they are situated.

I draw on a secondary analysis of qualitative longitudinal evidence on young people’s evolving expectations about accessing higher education in the context of significant higher education (H.E.) expansion in the UK; growing aspirations of young people with no familial background of university; and ongoing and sharp class-related inequalities in the chances of accessing university. A question arising is how young people’s expectations evolve over time, and if and how this varies by social class. The data were drawn from a single Timescapes project, Young Lives and Times (directed by Professor Bren Neale). The analysis was conducted by myself and a research fellow, Mandy Winterton. We came to this as secondary analysts: Mandy having had no role and Sarah a partial one in the primary project. In undertaking our analysis of longitudinal data here, we developed case profiles (cf. Thomson, 2007). We developed our longitudinal case-based analysis with reference to social diversity, exploring the interplay of specific influences on young participants’ educational identities and expectations about going to university, and how these evolved differently, through time, across social groups. We undertook a case-based analysis with reference to diversity within the qualitative sample as this mapped onto significant, class-related groupings across the population (for more details, see Winterton and Irwin, 2012). As secondary analysts, a sample structure may not be as we would wish it, so there was a need to maximize its potential. For example, YLT had a sample quite heavily weighted towards middle-class girls and a relatively high number of privately educated girls. The majority of the sample expected to go to university and we focused on this grouping in the analysis. There was, nevertheless, some diversity within this grouping (for example, in parents’ educational backgrounds). We read much of the available material, including longitudinal interview data from the ages of 14 to 17/18. We then selected for in-depth analysis a spread of cases chosen strategically to illuminate diversity in family background and resources. This spread was very revealing, also, of diversity in young people’s temporal experiences of family, school and peer influences in their evolving orientations to higher education (Winterton and Irwin, 2012). In this way then, we worked with an understanding of the sample structure and specificity, seeking to situate it with reference to wider evidence on inequalities and the ‘fit’ of the sample in relation to such evidence (in respect of both class pattern and salient processes).

The case-based approach helped us to explore the dynamics of social inequalities operating at a biographic, micro level of social experience, self-perception and interactions with significant others. Our analysis revealed the central salience of young people’s family (class related and higher education) background, an important dimension around which the data cleaved. We used this as an axis around which to organize further analysis of the linked influences of parental expectations as perceived by the young people, along with the influence of friends, school and teachers and other more contingent factors. In contexts where parents were in middle-class occupations and had been to university, it was also the case that parental expectations, school contexts and friendship influences were aligned, and pulled in the same direction. Here, youngsters held assured expectations of going to H.E. throughout their teenage years. Amongst youngsters expecting to go to university and with no family higher education background, there was an important division in the data. For some, expectations were vague at first but firmed up through their teenage years, in part through the influence of parental resources, expectations and investment in private education. Young people acquired a much firmer academic and prospective university student identity over time. Amongst other youngsters, there was greater contingency in their expectations. Here, parental expectations regarding university were ‘weakly framed’ (see Ball et al., 2002); and peer group and school-level influences tended to pull in different directions over time. Expectations were more subject to vagaries, and in a state of flux. Such vagaries were not random but had a logic and influence which was structured by circumstance and background. Overall, then, the evidence revealed the interplay of different influences over time and how these underpinned, or rendered uncertain, evolving ideas about going to university amongst youngsters from different backgrounds.

We suggest, then, that detailed longitudinal case-based analyses may orient us to the particular, but it simultaneously reveals how the interplay of factors over time varies by social background and circumstance and provides a revealing lens on the temporal, biographical confluence of processes shaping inequalities. A case-based longitudinal analysis organized with reference to how diverse (here class related) experiences were situated, and evolved over time, offered a powerful resource in theorizing the structuring of inequality. Whilst our resulting arguments about the shaping of diverse trajectories drew on a small sample, we sought to understand how its specificity related to extant evidence on patterns as well as processes, and to be clear about how our analysis could be tested in different contexts.

V Working across different qualitative datasets and issues of contextual specificity

Another strand of our secondary analysis activity entailed working with data from across the Timescapes projects (for example, Irwin et al. 2012 and Irwin and Winterton, 2012). Working across datasets puts into sharp relief some of the ways in which data are marked by the conditions of their production. Even where research questions had overlaps, very different project contexts meant it was far from straightforward to compare data directly. We worked quite closely with two of the Timescapes empirical projects where participants were parents of young children, specifically: ‘Work and Family Lives: The Changing Experience of “Young” Families’ (abbreviated to ‘Work and Family Lives’ and ‘WFL’) ⁴ ; and ‘Masculinities, Identities and Risk: Transition in the Lives of Men as Fathers’ (abbreviated to ‘Men as Fathers’ and ‘MaF’). ⁵ These both carried extensive data on identities and orientations to parenting. The Work and Family Lives project followed a set of sociological questions about work and family life with a particular interest in children’s as well as parents’ perspectives on managing time, and examined children’s, fathers’ and mothers’ experiences separately, as well as within family interviews. Men as Fathers was conceived with an interest in social and social–psychological questions, and men were interviewed about their own identities and experiences as these evolved from before the birth of their first child to when this child was eight. The different disciplinary and conceptual interests and the different sampling decisions and methods oriented participants to the projects in profoundly different ways. This presented questions about the appropriateness of treating data from across the projects as comparable, even where participants were talking about ostensibly similar aspects of their lives.

Secondary analysis often proceeds from engagement with one set of project data. For us, working across disparate projects engendered caution about seeing data as in any sense cumulative. There are examples of working across projects where analysts lay side by side data from different research projects. It may be that where differing projects provide commensurate evidence, it is due to sharing more tightly defined research questions than was the case for the diverse Timescapes projects. We proceeded by asking if we could enable a meaningful analytic conversation across datasets. We took as a focus, the area of gender and time stress in the family lives of parents with young, primary school-aged children. We built an understanding of its patterning with the Work and Family Lives project. Within this, we sought to explore experiences in more and less typical contexts, situating the data with reference to other extant evidence on gender and work–family conflict. Women across diverse circumstances appeared more likely to manage the work of work life balance than did men, and generally appeared more prone to experiencing pressure, particularly when they had extensive paid work commitments, even where a partner took on extensive practical caring work. We then undertook an analysis of Men as Fathers data, seeking to understand men’s perceptions and experiences of family and employment commitments, and exploring these across different typical and atypical contexts (Irwin and Winterton, 2014, forthcoming). In effect, we sought to bring evidence into comparison on the basis of translating our questions, and emergent hypotheses, to a new project context, as dissimilarity in project designs and samples meant that we could not simply ask identical questions across them. Rather we focused on the contexts in which time pressure was, and was not, experienced by women and men and brought these into comparison within, and across, projects. Whilst our efforts here were partial, they serve to illustrate a broader point: that secondary analysts need to be creative and critical in conceptualizing how to manage the specificity and contextual embeddedness of diverse datasets.

What lessons might be drawn for research in development studies? Qualitative researchers grapple with the question of how their analytic and conceptual claims can extend beyond the specificities of their study contexts. In part, it is the understanding of specificity itself which allows a situated understanding of the analytic reach of research. The approach we took in our own secondary analysis of data from across different projects entailed translating evidence between them so that we never let go of the specificity of study contexts. This cross-project component of our secondary analysis echoes the very different approach of meta-ethnography, a form of systematic review wherein researchers undertake an analysis of findings of multiple research projects. Analysts translate evidence between studies, seeking to remain true to the specific contexts in which data were generated whilst also identifying commonalities and developing a more general understanding of the phenomena under study (cf. Noblit and Hare, 1988). For example, Williamson et al. (2009) undertook a meta-ethnographic approach to analyzing constraints on contraception amongst young women in developing countries, drawing out commonalities across very diverse settings. Perhaps the most generalizable lesson I would wish to draw is that context counts. This will be old news to many development specialists, but it raises interesting questions relating to how secondary qualitative analyses might be undertaken. Secondary analysts need to be reflexive and critical, to challenge taken-for-granted categories and assumptions, to generate new questions, and they need to have a fine sensibility to the risks of being at a remove from the lived experiences represented in the data.

VI Conclusion

Using the example of secondary analysis of data from ESRC Timecapes, I have sought to highlight some issues relating to ethics and to the feasibility of effective secondary analysis, and to briefly explore conceptual issues relating to how data are situated and embedded in (and reflect) the contexts in which they are generated. These matters are not solely ones for secondary analysts, but addressing them was an important part of the analyses reported here, and also was integral to the approach of the Timescapes Secondary Analysis Project team in working with new datasets.

With respect to situating data, I took as one example an analysis of micro-level qualitative longitudinal data on young people’s evolving expectations about going to university. Whilst the available sample was marked by specificity in terms of its class and gender composition, it was possible to take advantage of key dimensions of diversity within the sample and conceptualize the data with reference to extant research and evidence on classed differences in how (family, school/teacher and peer group) processes intersect. The longitudinal evidence available then offered a picture both consistent with such evidence (adding confidence to an analysis based on a small sample) and provided additional insights into how such processes intersect over time, in class-varying ways, in young people’s biographies and evolving expectations.

It is a qualitative research commonplace to say that data are embedded in the contexts of their production, but it is also important to conceptualize the ways in which contexts (including research designs, disciplinary and methodological assumptions) are themselves embedded within primary data. Judgement about the contextually embedded nature of data should be part of evolving substantive analyses. Potentially, it can also contribute to productive strategies for working across datasets, and refining evolving concepts by bringing datasets into conversation and by appropriately translating evidence between them. In the example of gender and time pressure, I sought to illustrate scope for working across projects by drawing on data to address different, but commensurate, questions. The strategy is an example of how analysts might work with contextual specificity whilst still seeking to build insights across different data sources.

Throughout I have sought to draw out some themes which may have resonances for development researchers. As a non-specialist, whose secondary analysis practice is both recent and UK based, I find myself concerned that the ethical and epistemological difficulties may be marked, or even insuperable, in developing country contexts. The concern turns to anxiety when contemplating the damaging and sad history of ill-fitting external assumptions which have been imposed, even under the recent guise of progressive interventions (for example, Kar and Pasteur, 2005). It does seem plausible that secondary qualitative analysis might be used productively, where it is accompanied by a critical understanding of contextual specificities, that it might even be used to draw out such specificities and that it may engender new questions, new avenues of inquiry and, it is hoped, offer a helpful research resource.