Developing an Analytical Framework for Multi-Level Data Analysis in Qualitative Inductive Studies

Abstract

This paper proposes a multi-level framework for qualitative data analysis to address the challenges researchers face when integrating and synthesising findings from diverse data sources. While existing literature offers strategies for single-method qualitative analysis, few provide guidance on managing multiple data types within a unified framework. Drawing from an empirical example and relevant literature, the proposed framework presents a systematic approach that assists researchers in navigating complex data analysis, progressing from initial coding to theory development. The framework is structured into four levels: organising data, coding and generating themes, answering research questions, and theory construction. This stepwise process ensures both methodological rigour and flexibility, enabling researchers to iteratively explore relationships across datasets. Through an empirical example, we demonstrate how this framework facilitates effective triangulation, enhancing the credibility and coherence of findings. This approach is particularly beneficial for novice researchers, such as doctoral students, seeking to generate empirically grounded theories. The paper concludes with implications for future research and practical recommendations for applying the framework across various disciplines.

Introduction

Data analysis serves as a cornerstone in the research process. Increasingly, researchers are unsatisfied with a single data source and opt to collect a wide array of data types, triangulating them to reinforce and strengthen their findings (Creswell & Creswell, 2018; Denzin & Lincoln, 2013). This increases the complexity of analysing the data gathered far beyond that of analysing a single data source. While methodology textbooks abound in qualitative data analysis in general (e.g., Miles et al., 2020), there is limited methodological literature regarding the analysis of multiple data sources simultaneously and integrating them to generate a set of coherent findings (Zelčāne & Pipere, 2023). Moreover, researchers, particularly doctoral students, are increasingly expected to demonstrate their original contribution to the field by constructing substantive theories that are empirically grounded in the research findings (Higgins, 2007; Nagel et al., 2015). Our experience as former PhD students tells us that learning to analyse and theorise our findings is a challenging process fraught with insecurity, self-doubt and anxiety that is intensely stressful for the novice researcher. In recognising this gap, we propose a multi-level data analysis framework. We hope that the framework offers a systematic and methodical approach to help novice qualitative researchers better understand the data analysis and theorisation processes in the same way it has helped us. This paper is guided by the question:

How did we use multi-level qualitative inductive analysis to theorise research findings?

We examine this question within the scope of a research project that investigates the implementation of the International Baccalaureate Primary Years Programme (IBPYP) in relation to teaching intercultural understanding. The data collected included official IBPYP documents, teachers’ lesson plans, photographs of lessons and classroom settings, interviews, and an online survey with teachers.

In what follows, we provide an overview of existing data analysis methods, highlighting the strengths and weaknesses of each. Then, we present our proposed multiple-level data analysis framework, discuss the potential of multiple data sources, explain how the multi-level data analysis framework can be used to analyse various data sources, and demonstrate how it leads to theory development by illustrating the suggested approach with empirical examples from our research. We conclude by discussing challenges, implications and suggestions for future research.

Qualitative Data Analysis

In qualitative research, selecting appropriate data analysis methods is pivotal for determining research outcomes and insights. In our past empirical work, we have engaged with five distinct approaches: namely, thematic analysis, content analysis, grounded theory, narrative analysis, and discourse analysis (Berg & Lune, 2012; Bernard, 2013; Creswell, 2015). These methods enable researchers to organise, interpret, and draw meaningful conclusions from qualitative data (Creswell & Creswell, 2018). Each has offered valuable insights; however, when applied to projects involving multiple, diverse data sources, we encountered specific limitations that motivated the development of a more integrative analytical framework. Rather than claiming these approaches are universally dominant, we draw on them here to illustrate the types of challenges that arise when applying established methods to increasingly complex qualitative datasets. Together, they reflect a broad range of analytical traditions, from flexible descriptive techniques to structured theory-building strategies, offering a solid foundation for examining the need for a multi-level analysis framework.

Thematic Analysis is one of the most widely used and flexible methods for identifying and analysing patterns or themes within data. Its adaptability allows it to accommodate various research questions and datasets, making it a preferred choice in many qualitative studies. Braun and Clarke (2006) highlight its accessibility, particularly for researchers new to qualitative inquiry, but this very flexibility can also be a limitation. The method lacks the theoretical depth often needed for more specialised inquiries, which can result in surface-level findings that do not fully capture the complexity of the data (Barnatt et al., 2020; Jackson & Mazzei, 2012). Moreover, while thematic analysis was effective for identifying patterns in interview transcripts or single-source datasets, we found it less helpful in projects that involved multiple, heterogeneous data types, such as visual materials, observational field notes, and open-ended survey responses. In these cases, the framework offered limited guidance on how to integrate and interpret such varied sources coherently, prompting the need for additional tools or layered approaches.

Content Analysis is another commonly used method, known for its systematic approach to categorising and coding textual or visual data. Particularly suited for handling large datasets, content analysis allows for the quantitative analysis of qualitative data by focusing on frequency counts and the distribution of categories. Krippendorff (2004) emphasises its utility in media studies and communication research, where large amounts of text require organisation into meaningful categories. Qualitative content analysis encompasses a range of approaches. Some are primarily descriptive, focusing on categorising manifest content in a systematic way. In our empirical work, we employed this descriptive form of content analysis, focusing on IBPYP documents to identify curriculum emphases. By contrast, the qualitative content analysis described by Kuckartz and Rädiker (2023) integrates descriptive categorisation with reconstructive elements, enabling the interpretation of latent meanings alongside structural organisation. However, content analysis, particularly in its descriptive form, often emphasises surface-level meanings, which can lead to an oversimplification of data, as it may overlook the context and deeper meanings behind the text. In our previous studies, which involved open-ended interviews and reflexive fieldnotes, content analysis proved insufficient, as it did not adequately capture the evolving, situated nature of participants’ experiences. The method’s coding schemes were not easily adaptable to the ambiguity and fluidity inherent in these data types. This limitation became particularly apparent when working with diverse and complex datasets, prompting us to explore more flexible and layered analytical strategies.

Grounded Theory, developed by Glaser and Strauss (2017), offers a rigorous approach to generating theory directly from data. It is particularly valuable for exploratory research, where the aim is to build theory rather than test existing hypotheses. Grounded theory involves systematic coding and constant comparison, allowing for the emergence of concepts and categories that are grounded in the data. Bryant and Charmaz (2010) argue that its greatest strength lies in its ability to develop well-founded theories. However, the process is often time-consuming and requires intensive engagement with the data, making it less suitable for large-scale studies or those involving multiple data sources. Additionally, the inductive nature of grounded theory can complicate its application when datasets are diverse, as maintaining coherence across different types of data while allowing theory to emerge freely can be challenging (Holton & Walsh, 2017). In our own studies involving interviews, observational records, and institutional documents, we found that grounded theory’s coding procedures often prioritised the most narratively rich or theory-laden data (typically interview transcripts), which inadvertently marginalised other forms of data. This imbalance hindered our efforts to generate a theory that reflected the full complexity of the data sources in play.

Narrative Analysis takes a different approach, focusing on the stories told by research participants and the ways in which individuals construct meaning from their experiences. It is particularly useful in fields such as sociology, psychology, and education, where understanding the personal and collective experiences of participants is central to the research aim. Riessman (2008) notes that narrative analysis enables researchers to explore the richness of individual experiences and how stories reflect broader social and cultural contexts. However, its limitation lies in its specificity. While narrative analysis provides deep insights into individual perspectives, it struggles to generalise findings or integrate them with non-narrative data, such as quantitative findings or visual data. In our own work, we found that the emphasis on temporality, coherence, and plot structure in narrative analysis was challenging to apply uniformly across fragmented and structurally diverse data types, such as when combining interview narratives with observational fieldnotes or institutional documents. This limitation made it less adaptable for studies requiring the integration of diverse data sources.

Discourse Analysis is fundamentally rooted in linguistics, but different traditions engage with this linguistic foundation in different ways. Our focus is on the linguistic strand (Gee, 2014), which emphasises the textual and conversational aspects of meaning-making. Other traditions, such as the sociology of knowledge approach (Keller, 2011), extend discourse analysis to examine broader social and institutional structures. While discourse analysis proved useful for exploring micro-level linguistic meaning-making, it was less suited to our research projects that incorporated visual and numerical data, highlighting the need for a more integrative analytical framework. Denzin and Lincoln (2013) note that this method is highly interpretative, allowing for in-depth analysis of texts, conversations, and written documents. However, its abstract and interpretive nature can be seen as overly subjective, and the method often lacks the empirical grounding required for more straightforward, applied research. Although useful for exploring micro-level linguistic meaning-making, discourse analysis proved limiting for research that included numerical data, such as survey summaries. The method offered limited guidance for systematically analysing such heterogeneous data types, which reinforced the need for a more integrative analytical framework (Ezzy, 2013; Mertens, 2015).

Overall, each of these methods offers distinct advantages depending on the research questions and data involved. However, in our own research, which involved analysing interview transcripts, reflective journal entries, policy documents, and survey data, we found that these established methods each addressed only part of the analytical picture. This experience underscored the limitations of existing approaches when applied to multiple, diverse data sources. As qualitative research increasingly incorporates various forms of data, the need for more flexible and integrative analytical approaches becomes evident. To address this need, we asked: How can qualitative researchers rigorously and meaningfully integrate different types of data within a single study without losing analytic depth or coherence? Researchers must critically assess the scope of their data and research objectives when selecting a method to ensure it aligns with the complexity of the dataset and the depth of analysis required (Keane & Thornberg, 2024). This reflective process becomes particularly important when working with a multifaceted array of data types.

The Multi-Level Data Analysis Framework

Our proposed multi-level data analysis framework addresses the limitations of existing methods, which do not fully meet our needs. It enables us to integrate and synthesise findings from diverse data sources, a capability that is especially valuable when researchers aim to triangulate data to improve the trustworthiness and depth of their results. The framework is characterised by its flexibility, allowing it to be adapted to a wide range of research questions and data types. Thus, it offers a structured yet adaptable template for analysis, ensuring both methodological rigour and interpretative flexibility.

Additionally, the framework is designed with efficiency in mind, assisting researchers in managing the complexities of multi-level data analysis while minimising the time commitment often required by other methods. It also promotes the empirical grounding of new theories, aligning with the increasing expectations for PhD students and researchers to contribute original insights to their fields. Its capacity to accommodate both single and multiple sources of data facilitates theory building, systematises the analytical process, and builds progressively on findings from each source. Moreover, it illuminates nuances and interrelations within the data, enabling triangulation and corroboration, which in turn enhances the credibility and traceability of results. In summary, this innovative multi-level data analysis framework offers a comprehensive solution to the challenges posed by multiple data sources, making it a valuable addition to the toolkit of qualitative researchers.

Conducting Multi-Level Data Analysis on a Single Data Type

Qualitative data analysis is typically non-linear, iterative, and recursive (Creswell & Creswell, 2018). Researchers often provide examples of codes and claim that the themes ‘emerged’ from them without detailing a step-by-step analytical process. Frequently, the themes presented serve merely as labels for meta-categories or amount to little more than a method of organising areas. Novice researchers are then left wondering what is transcended in the data analysis process and how it contributes to theory development. However, we suggest that it is beneficial to break down the process into multiple levels, each building upon the previous level of analysis, although we acknowledge that these levels may overlap and are inherently flexible. In fact, some existing approaches operate with a multi-level or step-by-step concept. For example, building qualitative typologies (Kuckartz, 2010), conducting qualitative multi-level analysis (Hummrich & Kramer, 2018; Nohl, 2013), and performing the qualitative content analysis mentioned earlier. However, unlike Hummrich and Kramer (2018) and Nohl (2013), who use multiple levels in both theoretical and methodological contexts, we have used “multi-level” here to describe the analytical process involving several steps that need to take place in a specific order and relation to each other, rather than as “multi-level’ context research phenomena occurring at the micro, meso, and macro levels as described in Hummrich and Kramer (2018) and Nohl (2013).

This paper proposes a multi-level approach to qualitative data analysis, which was applied to our research to investigate the implementation of the IBPYP in relation to the teaching of intercultural understanding. Each level may include further micro-level analyses for different data types. The multi-level data analysis approach is particularly well-suited to the grounded theory methodology, where the ultimate goal of the researcher is to generate a substantive theory that explains the phenomenon under investigation based on insights from the data. In the following sections, we describe the phases of the data analysis.

Once the researcher is satisfied that they have collected all the necessary data for the project, it is crucial to begin sketching a framework to analyse the data, ensuring that all data sources are considered in the data analysis strategy. Figure 1 illustrates a data analysis framework within which the researchers utilise a four-level method: (1) organising the raw data, (2) generating categories and themes, (3) addressing the research question(s), and (4) theory development to analyse data in the interview transcripts and memos.

Figure 1.

Overview of a 4-Level Data Analysis Framework

Level 1: Organising the Raw Data

Qualitative fieldwork often generates data through interviews or focus groups, field notes, memos, and the collection of artefacts and documents, resulting in a considerable amount of data. The first step is to organise the data for analysis. This typically involves gathering and organising all the interview transcripts according to interview questions. This is an important step, as participants may not necessarily answer all the interview questions in a systematic and orderly manner. Hence, answers to the questions may overlap and be found in different sections of the transcripts. Hence, going through the transcripts in detail and aligning the sections with the respective interview questions is crucial. It is also important to note and archive any additional information provided by the participants that does not directly relate to the interview questions, as it may be helpful later in the analysis. We recommend keeping the original transcript intact in a separate folder and only working on the duplicate copy. After sorting the data, for example, according to the interview questions, the next step is to organise the data for cross-sectional analysis and comparison. This could involve further organising the interview responses according to participants’ profiles, for example, based on their gender, age, and the industries/sectors in which they work. This allows for comparing the data within and across the participant groups.

Level 2: Generating Categories and Themes

Once organised, the researcher needs to familiarise themselves with the data. This involves examining the data sources in detail and writing phrases that reflect first impressions that come to mind (Saldana, 2016). For interview transcripts, this may involve coding in a line-by-line, focused and thorough manner to avoid omitting any data at this preliminary stage (Charmaz, 2014; Urquhart, 2013). For artefacts, this may involve documenting important features, as well as keywords and phrases associated with the observation. These notes mark the earliest stage in the analytic process. Patton (1980) noted that “inductive analysis means that the patterns, themes, and categories of analysis come from the data; they emerge out of the data rather than being imposed on them prior to data collection and analysis” (p. 306). Thus, being open to new ideas and refraining from applying any preconceived notions at this stage of analysing the data is essential.

The process may yield numerous short descriptive phrases summarising the main discussion area in each section. In some instances, particularly for interview transcripts, this may also encompass in vivo codes—words or phrases taken directly from what the participants expressed (Birks & Mills, 2011; Glaser & Strauss, 2017). In vivo codes are regarded as symbolic markers of participants’ speech and meaning, although their usefulness should also be assessed with the same rigour as other codes (Charmaz, 2014).

Next, keywords and phrases that evoke similar emotions or describe identical situations are clustered. Those that relate to or contrast with one another analytically or conceptually are then grouped together. This process condenses them into conceptual categories. The next step involves generating themes. This requires the researcher to bring together categories that exhibit certain similarities, differences, and relationships, while assigning a phrase to each of these categories. These phrases are also commonly referred to as themes. It is essential that the data analysis is recursive and iterative to ensure that the themes are not hastily constructed.

Level 3: Answering Research Question

The generation of themes is an essential milestone in the data analysis process. The themes typically align with the study’s goals and, when considered in conjunction with the codes and categories, respond to the research questions. This level represents a critical juncture in the data analysis, where themes are interconnected to address the research questions. Depending on the research focus, participants may address the “final” research questions or provide answers to questions that may be linked to the research questions in a subsequent analytical step. Here, the researchers may decide whether to conclude their analysis at this practical level or delve into the more theoretical dimensions of their data. This shift towards theory development can lead to a deeper exploration of the research topic, providing a more comprehensive perspective and contributing to establishing theoretical frameworks within the field.

Level 4: Theory Development

“Theorising is the act of constructing … from data an explanatory scheme that systematically integrates various concepts through statements of relationship” (Strauss & Corbin, 1998, p. 25). As theories are “interpretations made from given perspectives as adopted or researched by researchers” (Strauss & Corbin, 1994, p. 279), the researcher’s prior experiences, knowledge and self-awareness contribute to the development of the theory (Birks & Mills, 2011). Charmaz (2014) describes theoretical sensitivity as “the ability to understand and define phenomena in abstract terms and to demonstrate abstract relationships between studied phenomena” (p. 161), and using theoretical coding in the later stage of data analysis helps to move the “analytical story in a theoretical direction” (Charmaz, 2014, p. 63). These theoretical codes are advanced abstractions that enhance the explanatory power of the findings and their potential as theories (Birks & Mills, 2011). To derive theoretical codes, it was necessary to think broadly and look beyond the researcher’s existing knowledge to identify keywords and concepts that might serve as potential theoretical codes to explain the analysis. Diagramming often helps identify relationships among categories and themes, including their strength, direction, and dynamics. This process contributes to theorising the findings. So, how do we determine that the analysis is sufficient and has reached the final data analysis phase? Lyn Richards (2014), in her book Handling Qualitative Data, suggests that substantive theorising is characterised by (1) Simplicity – a ‘small, polished gem of a theory’ rather than a ‘mere pebble of truism;’, (2) Elegance and balance – the theory is coherent; (3) Completeness – it explains all; (4) Robustness – it does not fall over with new data, and (5) It makes sense to relevant audiences. Theory development often represents the pinnacle of data analysis and the culmination of a rigorous data analysis process.

A Note on Memo Writing

Memo writing serves as a means to facilitate data analysis for some qualitative methods. Memos that were written in the early stages of the data analysis help explore emerging codes, categories, and relationships, while those written in the later stage of the research assist in placing the findings within an argument and making comparisons within and across categories and can therefore help to increase the level of abstraction of the data (Charmaz, 2014). Memo writing serves many purposes. It encourages the researcher to engage in critical reflexivity by making the researcher’s standpoint and assumptions visible: it provides a space for the researcher to actively engage with the materials by allowing comparison between and across data, codes, categories, and themes. This process facilitates conceptual abstraction by assisting in the transition from descriptive initial codes to conceptual categories and, subsequently, to theory construction (Charmaz, 2014).

Conducting a Multi-Level Data Analysis on Multiple Data Types

Researchers increasingly commonly collect data from multiple sources to ensure that the research topic is approached from diverse research contexts. Combining different data sources can provide a more comprehensive and holistic view of the topic and allow for data triangulation and corroboration (Keane & Thornberg, 2024). The strengths of one datum can compensate for the weaknesses in another, enhancing the validity and reliability of the results. In some cases, the only option may be to combine sources to gather all the necessary information about the research object.

Analysing multiple data sources and synthesising the results can involve various data sources, such as interviews, observations, documents, images, or videos, and different methods of analysis, including coding, thematic analysis, narrative analysis, or discourse analysis. The questions are: How can one effectively integrate and synthesise data from multiple sources in qualitative data analysis? How can one develop a theory grounded in these multiple data sources?

Combining data from multiple perspectives further amplifies the complexity of analysing the data gathered, surpassing the initial challenges posed by each individual data analysis approach. Recognising these difficulties, we argue that complex and multidimensional data require innovative strategies for data analysis (Holland et al., 2006). To fully harness the potential of such data, a systematic and methodologically sound approach is essential for analysis. Firstly, to enhance the credibility and trustworthiness of qualitative research (Denzin & Lincoln, 2005), and secondly, to help make the complexity of analysing multiple data sources more manageable.

The potential of analysing multiple data sources in a multi-level format lies in its capacity to facilitate comparisons within and across datasets. This not only yields insights into the multiple perspectives within each type of data but also highlights the nuances and possible biases inherent in each dataset, which are critical for the validity and credibility of the findings.

In the following section, we illustrate how we use the multi-level data analysis framework in our project. The project research question was: “How do teachers implement the IBPYP to develop primary students’ intercultural understanding?”. An illustration of the data analysis process is shown in Figure 2.

Figure 2.

Using Multi-Level Data Analysis on Multiple Data Sources

Level 1: Data Collection, Organisation, and Initial Coding

At the initial level of data collection and organisation, the research study drew from various sources to comprehensively explore the topic of teaching intercultural understanding in the IBPYP. One critical source was document analysis, which involved examining official IBPYP documents and guidelines related to intercultural understanding. Additionally, primary data were collected through semi-structured interviews with IBPYP teachers, aiming to gather teachers’ perspectives, experiences, and practices related to teaching intercultural understanding. Furthermore, an online survey was administered to a broader sample of teachers to collect quantitative data that complemented the qualitative insights gained from the interviews.

Once the documents were collected, the data organisation phase came into play. During this stage, the researchers systematically organised and categorised the information obtained from the document analysis. This step laid the groundwork for the subsequent stages of analysis by providing a structured overview of the program’s official stance on this topic.

Through initial coding, key themes and concepts related to intercultural understanding within the IBPYP were identified and structured. Simultaneously, the researchers conducted an initial analysis of the semi-structured interviews and survey data, further enriching the dataset. These interviews and survey responses offered a comprehensive understanding of how IBPYP teachers perceived and practiced intercultural understanding in their classrooms.

Level 2: Focused Coding, Category Formation, Theme Development

With data collected from interviews and lesson plans, the analysis progresses to the second level, characterised by focused coding. This entails organising and coding the interview transcripts, pinpointing recurring themes, concepts, and patterns present in teachers’ responses. Through this process, researchers begin to identify and develop preliminary themes that emerge from the rich qualitative data. Simultaneously, the lesson plans are carefully examined, and categories are created based on the various approaches and strategies employed by teachers to teach intercultural understanding. These categories reflect the specific methods and content used by teachers in their classroom practices.

Moving further, the categories derived from the interview data are used to form broader categories that encapsulate different aspects of intercultural understanding as perceived by the teachers. Similarly, the categories generated from lesson plan analysis are structured to align with these overarching themes. As the analysis unfolds, researchers engage in a deeper exploration of these emerging themes, seeking to understand the underlying concepts and connections within the data. This level of analysis enables researchers to identify both the specific details and the higher-order concepts related to intercultural understanding in the IBPYP context, ultimately contributing to a more comprehensive understanding of the research topic.

Level 3: Comparative Analysis and Answering the Research Question

At Level 3, the focus shifted towards comparative analysis and addressing the research question. During this phase, researchers scrutinised the themes and categories derived from the interviews and the analysis of lesson plans. This detailed examination allowed them to identify both convergences and discrepancies between teachers’ perceptions and their instructional practices. This stage provided valuable insights into how teachers’ beliefs aligned with their actions in the classroom, uncovering the complex interplay between pedagogical intentions and implementation. These revelations were achieved through the comparative analysis across different datasets.

Level 4: Theory Development, Conclusion and Implications

The last level focuses on theory development, drawing conclusions, and presenting research output. This phase was dedicated to further refining and articulating the theoretical framework. The researchers drew upon their comprehensive analysis, incorporating insights from existing theories, to develop a theoretical model that encapsulated the complexities of intercultural understanding in the IBPYP context, having considered various dimensions and variables.

Discussion

The multi-level data analysis framework presented in this article offers several notable strengths. It provides a systematic, methodologically sound approach for analysing complex, multi-dimensional data from various sources. This is particularly crucial as qualitative research increasingly involves multiple data types, including interviews, documents, images, and quantitative data. This reflects broader discourses in mixed methods research, which emphasise the value of integrating qualitative and quantitative data to enhance analytical richness and validity (Creswell & Plano Clark, 2018). By integrating and synthesising such data, the framework enhances the credibility and trustworthiness of qualitative research, contributing to research validity. The ability to cross-reference and corroborate findings across different data sources strengthens the depth of analysis and minimises potential biases, allowing researchers to uncover patterns and nuances that might otherwise be overlooked in single-method approaches.

Strengths of the Multi-Level Data Analysis Framework

One of the most significant strengths of the multi-level framework is its capacity for cross-comparative analysis. The framework’s ability to compare and contrast data from various sources offers a more nuanced understanding of complex social phenomena. For instance, when different data types (e.g., interview transcripts and documentary evidence) yield conflicting results, the framework enables the exploration of these divergences in a structured manner, assisting researchers in identifying potential biases, contradictions, or contextual influences. This integration of multiple perspectives offers a richer and more comprehensive understanding of the research problem (Herman & Parsley, 2022; Repko & Szostak, 2021).

Additionally, the framework provides flexibility in methodological application. While it offers a structured template, researchers can adapt the framework to meet the specific needs of their study. This flexibility is particularly valuable in fields that address multi-layered phenomena, such as education, healthcare, or social policy, where data often originate from diverse and sometimes conflicting sources (Repko & Szostak, 2021).

The framework facilitates the incorporation of multiple voices and perspectives, promoting a holistic approach that is increasingly expected in contemporary qualitative research. Another strength lies in the framework’s alignment with the growing emphasis on rigorous theory-building within qualitative research. By progressing through various levels of analysis—organising data, coding, generating themes, and developing theory—the framework offers a clear and systematic pathway from the initial data corpus to theory construction. This process underpins the empirical grounding of new theories, which is particularly relevant in fields such as sociology, education, and social sciences, where researchers are expected to contribute original, theory-driven insights (Ashworth et al., 2021). In particular, the framework is well-suited for researchers employing grounded theory, where the goal is to develop theories closely tied to data.

Broader Applications Across Disciplines

The framework’s potential is not limited to a single field; it can be applied across various disciplines where multiple data sources and complex datasets are commonplace. In educational research, for instance, the framework can facilitate the integration of classroom observations, student performance data, and teacher interviews, offering a more nuanced understanding of how pedagogical practices impact learning outcomes (Cohen et al., 2018). Similarly, in healthcare research, the framework enables the synthesis of patient interviews, clinical data, and observational records, thereby contributing to a more comprehensive understanding of patient care practices and treatment outcomes (Dixon-Woods, 2011).

The framework is particularly valuable in interdisciplinary studies, where researchers often face the challenge of integrating data from distinct methodological traditions (van Baalen & Boon, 2024). For example, in environmental research, data from scientific measurements, interviews with local communities, and policy documents can be systematically combined to provide a multi-dimensional analysis of environmental challenges (Creswell & Plano Clark, 2018). This capacity for interdisciplinary application further highlights the versatility and utility of the framework in contemporary research landscapes, where complexity and interconnectedness are increasingly emphasised.

Challenges and Limitations

Despite its strengths, the multi-level data analysis framework comes with specific challenges and limitations. One of the primary challenges is the time-intensive nature of the framework. Organising, coding, and analysing multi-source data requires significant time and effort, particularly when dealing with large or highly complex datasets. Researchers must be prepared for a potentially lengthy analysis phase, which can be a drawback in time-constrained research projects. Additionally, the framework’s complexity demands a high level of methodological expertise, particularly in using coding techniques and qualitative data analysis tools. For novice researchers, the learning curve associated with this framework may be steep, necessitating training in data management and analysis. Another potential limitation is that while the framework provides flexibility, it may not be suitable for all types of research questions. Researchers must critically assess the scope of their study and determine whether the framework’s multi-level structure aligns with their specific objectives. Not all research questions require multi-source data analysis, and in some cases, more streamlined methods (such as thematic analysis or basic content analysis) might be more efficient and effective.

Practical Considerations for Implementation

Successful implementation of the multi-level data analysis framework requires careful planning and organisation. Researchers must begin with a clear understanding of their research questions and the types of data they will be analysing. Integrating diverse data sources demands meticulous attention to detail in terms of coding and categorisation. Qualitative data analysis software, such as NVivo or Atlas.ti, can facilitate this process by allowing researchers to organise, code, and manage large datasets efficiently. However, researchers must be familiar with these tools to fully leverage their capabilities.

Maintaining transparency throughout the data analysis process is another crucial factor for ensuring the reliability and replicability of findings. Researchers should document each step of the analysis, including how coding decisions were made, how themes were developed, and how theory was constructed. This transparency is essential for enhancing the credibility of the research, as it allows other researchers to follow the analytical process and potentially replicate the study in different contexts. Finally, the iterative nature of the framework means that researchers must be prepared to revisit earlier stages of the analysis as new insights emerge. This cyclical process of constant comparison and refinement can be time-consuming but is essential for ensuring that the final theoretical framework is robust and grounded in the data.

Conclusion

This article proposes a multi-level data analysis framework as a valuable tool to support researchers in navigating the intricacies of data analysis for single and multiple data sources. Using an empirical example derived from one of the authors’ doctoral research projects, which investigates the teaching of intercultural understanding within the IBPYP, they illustrate how they utilised the multi-level data analysis framework to facilitate theory development. The framework’s strengths lie in its ability to enhance data credibility, provide a systematic approach, and enable comparative analysis. While careful consideration and documentation are necessary, their practical implications for research and theory development are substantial. Researchers across diverse disciplines can leverage this framework to address complex research questions and contribute meaningfully to their fields.