Socio-Technical Grounded Theory
for Software Engineering

Software engineering abounds with socio-technical phenomenon, where human and technological interactions are tightly coupled, such that studying one without the other makes for an incomplete investigation and understanding. For example, the core SE practice of programming is not strictly technical, rather, it is socio-technical, involving intensive human-human collaboration and coordination and human-technology interactions. Similarly, most SE practices such as pair programming, designing prototypes, customer collaboration, software deployment, maintenance and DevOps, are all socio-technical in nature.

Research communities such as human-computer interaction (HCI), computer-supported cooperative work (CSCW), and cooperative and human aspects of software engineering (CHASE) have been at the forefront of exploring the human and social aspects. Most topics covered by these areas involve socio-technical, rather than strictly social, phenomena. More recently, with the rise of artificial intelligence (AI), there is renewed interest in gaining in-depth understanding of human factors (Hidellaarachchi et al., 2021), human values (Perera et al., 2020; Hussain et al., 2020), and the human-in-the-loop in order to build more realistic, responsible, and usable AI-enabled systems.

Some contemporary phenomena and future topics that are particularly suited for STGT studies in SE/AI and other socio-technical domains include:

• •

  understanding human aspects (e.g. personality, gender, age, emotions, motivation, etc.) in SE/AI,

• •

  understanding human values (e.g. ethics, privacy, safety, security, morality, etc.) in SE/AI,

• •

  understanding social aspects (e.g. collaboration, coordination, formation etc.) in SE/AI,

• •

  role of technology (e.g. social media, virtual/remote collaboration tools) in crisis management, remote work, online education, political campaigns, gaming, etc.,

• •

  human- and social-centered design of inclusive and responsible SE/AI systems,

• •

  understanding socio-technical aspects of specific SE practices, e.g. requirements elicitation, user-centered design, estimation, pair/mob programming, etc.

Acknowledging these phenomena as socio-technical, in SE and other domains, is the first step to enabling more complete and comprehensive interpretations, constructions, and rendering of modern human realities.

With increasing proliferation of technology and digitalisation, most domains are becoming socio-technical to varying degrees, e.g. banking, medicine, education, and retail. Domains such as SE, computer science, human computer interaction (HCI), artificial intelligence, and information systems, are inherently and inextricably socio-technical domains, with tight coupling between its social and technical aspects. Domains such as SE do not simply represent a user base of technology, they are the birthing cradle of information technology and systems. Technology is not only a prominent enabler or feature in these domains, it is the core raison d’être for organisations, e.g. for software companies.

Actors, people who play key roles in the phenomenon, in these domains are not regular users of technology. For example, software engineers, data scientists, computer scientists – the actors in the SE domain – are the creators and maintainers of software technology, tools, and platforms that are increasingly becoming indispensable to all domains. Often referred to as ‘geeks’, they display their own unique language, interactions, culture, worldviews, and social norms.

Software engineers are particularly interesting to study, as they are both the producers and users of technology. They are not only the architects of the software platforms that enable multiple realities, they too function and interact within combinations of these worlds. For example, using a combination of real and digital workspaces, artefacts, and communication mechanisms such as GitHub, physical Scrum task boards, online tools such as Trello or JIRA, physical and virtual pair programming, using Slack channels for team communication, face-to-face and Zoom meetings. All these involve intensively socio-technical interactions.

More generally, depending on the domain and the phenomenon, the actors being studied may be software engineers including developers, business analysts, testers, senior managers, or software users, including classroom teachers, spacecraft designers, satellite developers, government officials, doctors, nurses, psychologists, artists. Additionally, STGT studies may involve actors that are human, virtual avatars, or AI agents, depending on the nature of the realities where phenomena occur and are studied.

A socio-technical researcher has the requisite knowledge and skills from social sciences, philosophy, qualitative research, and theory development as well as technical background and/or experience applicable to the domain. Lack of sociological background is seen as hampering grounded theories (Martin, 2019; Gibson and Hartman, 2013). This does not imply that an ST researcher needs to be an expert in each of these areas. Neither do they have to be one person. To conduct quality STGT studies, a socio-technical researcher, or an interdisciplinary team of researchers, should typically possess a core repertoire of knowledge and skills.

• •

  philosophical foundations – understanding fundamental philosophical concepts such as types of reasoning, ontology, epistemology, and research paradigm. A basic overview of these concepts is presented in section 4.2.

• •

  qualitative research – designing research protocols, ethics assessment/approval, collecting and analysing qualitative data.

• •

  theoretical sensitivity – affinity for theoretical abstraction, conceptualisation, and theory development.

• •

  socio-technical sensitivity – an affinity for exploring and understanding the intricacies of human interactions and socio-technical aspects of phenomena.

• •

  domain knowledge – an understanding of the research domain, typically provided by a subject matter expert.

When equipped with relevant background in social sciences and theory development, SE researchers become socio-technical researchers. Those successful in conducting quality GT studies in SE so far (Table I) have applied their repertoire of knowledge and skills gained from social sciences, philosophy, theory development, and SE. Researchers lacking the requisite knowledge and skills struggle to achieve quality research outcomes (Storey et al., 2020).

Similarly, sociologists with required technical background are well placed to be socio-technical researchers, conducting STGT studies of phenomenon with prominent and interwoven role of technology. An inter-disciplinary team of software engineering researchers or technologists, sociologists, and domain experts (e.g. banking and finance researchers for studies in the banking domain) can also make for an effective cross-functional socio-technical research team to study an increasingly socio-technical world. The importance of inter-disciplinary and trans-disciplinary research teams is not only relevant to STGT studies, but is also being acknowledged in empirical software engineering in general (Fernández and Passoth, 2019).

Socio-technical researchers studying socio-technical phenomena, domains, and actors can leverage from a range of traditional and modern research data, tools, and techniques.

Traditional data types, sources and collection techniques such as interviews, observations, and field trips offer rich data and insights. However, since the introduction of GT about half a century ago, a number of modern research data, tools, and techniques have emerged. While traditional GT guidelines are not averse or impervious to them, they understandably do not provide explicit guidance or examples of how best to harness them.

SE researchers are at the forefront of the technological revolution in research. Modern research data, tools, and techniques have given rise to entirely new research communities, e.g. mining software repositories. Modern types and sources of data such as publicly available software code repositories with commit messages, source code, documentation, and wikis regularly used in SE research, make for potent new avenues for theory development. Especially, when combined with modern techniques such as data mining and natural language processing, they can be used to extract large amounts of data, hitherto impossible with manual techniques. However, easy access to publicly available datasets and powerful search tools can be tempting to practice ‘data-driven research’, where easy access to large datasets precedes and motivates, often trivial, research questions and studies. On the contrary, the use of publicly available, large datasets in STGT is envisioned as a mechanism to scale theories to cover wider contexts of use, while still being firmly grounded in empirical evidence and following systematic and rigorous STGT steps and procedures to iteratively and incrementally derive key patterns from underlying data.

Additional sources of data such as research literature and grey literature are also being explored for theory development (Martin, 2019). In fact, Wolfswinkel et al. (2013) have presented a GT-based approach to conducting literature reviews, called the grounded theory literature review (GTLR). Future STGT studies in SE can explore these avenues to create modern renditions of grounded theories.

Furthermore, traditional GT analysis techniques, open coding and constant comparison, can be complemented with modern analysis techniques such as sentiment analysis (Calefato et al., 2018), e.g. to study emotions in software engineering. Currently, such innovations are limited to individual studies (Madampe et al., 2020). In order to influence improvements in GT practice in SE research at scale, the use of modern data, tools, and techniques need to be acknowledged and documented as method guidelines.

Future STGT studies can explore the use of cutting edge tools and techniques such as artificial, virtual, augmented, and mixed reality devices and platforms to achieve immersive and experiential data collection in natural and alternative realities. The improving efficacy of automation tools promises to ease and augment many research steps such as transcriptions, data collection, and even analysis.

GT challenged the top-down hypothetico-deductive approach to verifying existing theories dominating the sociological research scene in the mid-1960’s and steered it toward a bottom-up inductive approach to generating new and original theories. In a bid to delineate its stance, traditional GT overemphasises the role of induction but classifying GT as purely inductive is naïve (Haig, 1995). Different types of reasoning, inductive, deductive, and abductive are seen to apply in theory development. Please refer to Table III for their simple definitions and examples.

The role of deduction in guiding theoretical sampling is acknowledged in Glaserian GT (Glaser and Strauss, 2017, chapter II). Strauss-Corbinian GT suggests a greater role of deduction in the later stages of a GT study as the emergent theory is systematically verified against the data (Strauss and Corbin, 1990) while Glaserian GT is strongly opposed to this approach, claiming it as “forcing” rather than emergence of theory (Glaser, 1992).

On the other hand, the role of sudden or slow-dawning “sensitive insights of the observer” is acknowledged as “the root sources of all significant theorizing” (Glaser and Strauss, 2017, chapter XI). While the descriptions of these “insights” are close to the definition of abduction by Charles Sanders Peirce who defines it as “a process of forming an explanatory hypothesis” (Douven, 2017; Haig, 1995), Glaserian GT does not explicitly acknowledge the role of abduction in theory development. Strauss-Corbinian GT is said to have been influenced by abduction but its reference remains limited to a footnote in Strauss’ book (Charmaz, 2014; Strauss, 1987).

Charmaz acknowledges abduction as a type of “inferential leap” required of the GT researcher when they arrive at a surprising finding that does not fit the emergent patterns (Richardson and Kramer, 2006; Charmaz, 2014). Using this “imaginative way” of reasoning, the researcher comes up with useful explanations (theoretical conjectures or inferences) in an attempt to account for the surprising finding, then returns to re-examine or collect more data to check them (Charmaz, 2014).

The role of abduction in developing grounded theories remains to be fully explained (Bruscaglioni, 2016; Martin, 2019). Understanding different abductive reasoning modes such as reason or hunch, clue, metaphor or analogy, symptom, pattern, and explanation (Shank, 1998) opens a variety of avenues for creative thinking which is core to theorising and theory development (Martin, 2019). For example, abduction is the basis for diagnosis in medical sciences, where plausible hypotheses are explored and then accepted or discarded based on how well they ‘fit’, i.e. explain the symptoms being observed. Criminal investigations also offer examples of the three types of reasoning, particularly abduction in action, most prominently demonstrated in the engaging pursuits the fictional character of Sherlock Holmes, but also referred to by Peirce from a personal ‘detective’ experience (Reichertz, 2007). Similarly, abduction is applied in research for theory development to suggest hypotheses based on evidence. Through iterative data collection, analysis, and memoing, these hypotheses fall through or become established over the course of the study. Additionally, coming up with robust hypotheses about a socio-technical phenomenon requires a full understanding of the socio-technical context.

STGT, like traditional GT, follows a predominantly and overarching inductive approach, drawing generalised conclusions from patterns evidenced across specific instances. Induction in STGT is most clearly manifested in its open and targeted coding and constant comparison procedures where evidence is repeatedly raised in levels of abstraction toward a theory. STGT acknowledges and explains the role of deduction as applied in theoretical sampling and, unlike traditional GT (see Table IV), the role of abduction as applied in memoing, theoretical structuring, and theoretical integration (procedures described later.)

Ontology refers to what we believe exists or what we perceive as reality, in a research context. As researchers, we hold different views about what constitutes reality in the research context, referred to as our ontological stance. For example, we may believe only directly observable material bodies in the natural world constitute reality that is certain or that reality includes socially constructed concepts such as identity, culture, and beliefs, and combinations thereof. Researchers inclined to study the former (material reality) are often seen to prefer what are typically referred to as ‘hard’ sciences such as physics and maths, where the reality of the research context includes relatively stable phenomena (e.g. acceleration, gravity) studied through observing or measuring actions on interchangeable objects (e.g. apples), typically using quantitative research (numbers, accuracy, precision). Researchers inclined to study the latter (socially constructed concepts as reality) are often seen to prefer what are traditionally referred to as ‘soft’ sciences such as psychology and sociology, where the reality of the research context includes phenomena that are full of variations (e.g. driving, cooking, programming) studied through interactions with non-interchangeable subjects (e.g. people), typically using qualitative research.

Software engineering research predominantly involves studying socio-technical phenomena that unfold in multiple realities, beyond the traditional physical reality, e.g. in artificial, virtual, augmented, and combinations of realities. For example, studying a typical software development team involves physical contexts (e.g. physical scrum boards) and physical interactions (e.g. physical daily standups) as well as virtual contexts (e.g. project management on JIRA or Trello), virtual interactions (e.g. video calls, emails, virtual chat, online discussion forums), and even combined interactions (e.g. two developers collaborating in-person over a virtual task board self-assigning tasks using symbols such as their online avatars). Contexts such as remote work are primarily virtual, offering different affordances, norms, and symbolism (e.g. “you are on mute” makes no sense in a physical meeting). Studying social interactions between real people using their online profiles and avatars on online social media platforms such as Twitter and Facebook presents yet another type of reality as real versus fake identities come into play (e.g. symbols such as the ‘blue tick’ on Twitter enabling authentication.) These multiple realities are defined by intermixing operational contexts, symbols, languages, norms, and guidelines of both physical and virtual worlds, offering same if not more complexity than the physical world. While traditional GT guidelines address physical realities, STGT acknowledges the diverse and combined ontological frames prevalent in SE and provides guidelines to enable the study of physical, virtual, combined, and simulated realities.

Epistemology refers to what can be treated as knowledge and how that knowledge is gained, in a research context. As researchers, we hold different views about how we can gain knowledge about the reality of our research context, referred to as our epistemological stance. For example, we may believe that we can learn about the research world (objects and phenomena therein) through neutral measurements or observations where exchanging an observer with another will yield the same results. This is referred to as an objectivist epistemology. On the other hand, we may believe that we can learn about the research world through subjective interpretations of observations where exchanging the observer with another will likely yield different findings. This is referred to as a subjectivist epistemology.

STGT highlights that a researcher’s epistemological stance is typically determined by two aspects, the nature of the reality being studied, e.g. physical, virtual, augmented, combined, and the researcher’s own preferences. For example, STGT studies of virtual game worlds can be designed with an objectivist approach where the game analytics are used to collect objective metrics about the game (e.g. lives lost, levels successfully cleared, time taken per level, choice of actions and interactions). Similarly, STGT studies of combined realities, such as development team environments, can apply a mix of objective metrics and observations that are interpreted subjectively.

A combination of ontological and epistemological stances form a research worldview. Research paradigms formally capture the researcher worldview about what they believe is reality (ontology) and how knowledge about that reality can be gained (epistemology), in a research context. Popular paradigms include positivism, post-positivism, constructivism, symbolic interactionism, and pragmatism. Let us consider the prominent paradigms associated with GT below, compared under philosophical foundations in Table IV.

Positivism is naturally associated with quantitative data and deductive reasoning, drawing specific conclusions from general facts through repeated testing and proving/falsification of hypotheses. Historically, quantitative researchers with positivist inclinations have been staunch critics of qualitative research often conducted with interpretive approaches (Charmaz, 2014). Post positivism, on the other hand, refocuses attention on falsification of hypothesis, increasing confidence in proposed theories with every failed attempt at falsification (Easterbrook et al., 2008).

Classic or Glaserian GT has been associated with a positivist approach (Charmaz, 2006).

Constructivists emphasise generation of new knowledge or theories over verifying existing knowledge or theories, and value originality, novelty, usefulness, and real-world significance. Research methods most closely aligned with a constructivist approach include ethnography and case studies (Easterbrook et al., 2008).

The primary premise of introducing a new version of GT by Charmaz was to introduce and apply a constructivist research approach to conducting GT (Charmaz, 2006).

In addition to research paradigms, some unique perspectives can be applied to research. Feminism is concerned with the representation of females in the studied phenomena and so applies that lens to all aspects of research including research design, data collection, analysis, presentation, and evaluation. Critical theory is based on the belief that research serves a political purpose and can be used to empower specific groups, specially minority groups through raising awareness and recommending change.

While no one research paradigm or worldview is correct or wrong, it is important for researchers to be aware of and declare their philosophical stance because it influences how we conduct studies, present findings, and most prominently, how we evaluate research. For example, a positivist reviewer evaluating a research study conducted using a constructivist approach is likely to – consciously or unconsciously – apply positivist evaluation criteria, inappropriate for the study, e.g. look for generalisability instead of novelty or expect reproducibility and replicability instead of credibility and rigour, and vice-versa. Further information on research paradigms and their impact on research design can be consulted in (Seaman, 1999; Easterbrook et al., 2008; Wohlin and Aurum, 2015; Russo and Nuseibeh, 2001). The use of research paradigms in SE research has been explored in (Melegati and Wang, 2021).

An interesting point to note is that an individual’s approach to life in general, i.e. their general worldview, while likely to influence their research worldview, need not be the same. For example, an individual with a pragmatic approach to making their life choices may consciously chose to apply a positivist approach to their research study.

More often, researchers prefer a research paradigm over another and resort to applying their preferred paradigm for a majority of their research studies. However, researchers need not be bound to any one research paradigm for all their studies and can chose to conduct different studies using paradigms best suited to the nature of the study. Such flexibility requires a mature understanding of philosophical foundations and high levels of researcher self-awareness and reflection. Similarly, research paradigms and perspectives can be combined, e.g. critical positivist or feminist-constructivist to achieve unique research aims and flavours.

Traditional GT methods have been associated with particular paradigms, Glaserian GT with positivism (Charmaz, 2006), Strauss-Corbinian as building on pragmatism and symbolic interactionism (Strauss and Corbin, 1990), and Charmaz’s GT with constructivism (Charmaz, 2006).

STGT, as the name suggests, applies a socio-technical worldview to conducting GT research, which is open to selecting and applying specific ontological and epistemological stances as best suited to the research context. In other words, STGT can be applied using different research worldviews based on the ontology applicable to the study and on the researcher’s preference. This does not equate to being paradigm ignorant or agnostic, rather it requires a careful consideration of the research context to ascertain the choice of paradigm to apply.

For novice researchers, some suggestions are provided. For example, STGT studies of physical and/or combined realities, such as offered by software team environments, may be conducted using a subjective approach where the nature of the findings are dependent on the researcher conducting the data collection and analysis. On the other hand, STGT studies of simulated and virtual realities, such as in computer games, may apply an objective approach, leveraging in-game analytics. Studying end-user interaction with AI manifested as chatbots or robots offers similar contexts. Such simulated and combined realities can be designed to be finite and deterministic, manifesting closely the idea of an external standalone reality with limited possible interpretations, best captured by a objectivist stance.

Some research areas can be studied using different worldviews depending on the study focus and objectives. For example, a study on test case selection can be conducted using an objective approach if focusing purely on the technical tools and their performance, validating and verifying outcomes against goals. While a constructivist approach can be applied to the same research context if considering the tester’s motivations, rationales, and behaviours with regards to test case selection. STGT studies can also combine a selected epistemology with established theories and perspectives used as an additional lens, e.g. a constructivist-feminist STGT study of female contributors to open source software, their experiences, challenges, and strategies.

By acknowledging the research paradigms and perspectives, researchers can remain aware and consistent in their application, and actively set the expectations for their readers and reviewers so that the findings can be understood and evaluated in commensurate ways.

STGT supports different sampling techniques to get data collection underway, e.g. convenience, purposive, random, or representative sampling, as applicable to the project context and constraints. Once the iterations of basic data collection and analysis start yielding concepts and categories, theoretical sampling can be employed. Theoretical sampling is the ongoing process of assessing the emerging codes, concepts, (sub)categories, and hypotheses, and targeting specific data sources and types for collection in the upcoming iterations that are likely to help identify, develop, and saturate them while filling any theoretical gaps.

Theoretical sampling has been described a “complex form of sampling” (Coyne, 1997). While not including theoretical sampling explicitly, traditional views classify all sampling in qualitative research as purposeful, intentionally selecting data sources based on their specific characteristics (Patton, 1990; Sandelowski, 1995). In this sense, theoretical sampling has been classified as purposeful (Coyne, 1997), driven by the emerging theory, while its dynamic nature can also lend itself to the snowballing strategy. A distinguishing feature of theoretical sampling is that it is not pre-determined, rather ongoing and drives interleaved data collection and analysis.

Because it involves assessing high-level findings to direct data collection in specific instances (moving from general to specific), theoretical sampling can be said to apply deductive reasoning (see Table III) (Becker, 1993). For example, applying theoretical sampling to the emergent self-organising roles on agile teams based on data collected from relatively new agile teams in the early stages of the study (Hoda et al., 2010) led to identifying theoretical gaps around the maturity of the roles. The later parts of the study employed theoretical sampling to target data collection from mature agile teams leading to the subsequent mature and saturated definitions of these roles (Hoda et al., 2012b). Theoretical sampling includes grooming and refinement of the research protocols from time to time, e.g. refining interview questions or keywords for mining public repositories to progressively focus on emerging concepts.

Being aware of the full socio-technical context of the study domain will mean researchers can apply sampling effectively to identify data sources such as participants (e.g. software developers, users, stakeholders) and documentation (e.g. app reviews, GitHub wikis) and apply appropriate data collection techniques such as physical or virtual interviews and observations and automated or manual mining of software repositories and online forums.

STGT accepts a variety of data sources as well as collection approaches. The modern socio-technical world with its multiple realities offers new public sources (e.g. publicly available practitioner blogs, presentations and talks as YouTube videos, and opinion trends on Twitter) and domain-specific sources (e.g. digital project management boards such as Trello, open source software (OSS) repositories such as GitHub, communication channels such as Slack) and modern collection approaches (e.g. immersive and experiential approaches on online social media or in virtual game worlds.)

Data collected from traditional sources such as semi-structured interviews and observations continue to be important, providing rich information, facial and voice cues, and opportunities for follow-up questioning. They offer real examples verifying participant statements and adding necessary contextual background, but are limited in scale due to the manual effort required. Modern data publicly available, on the other hand, e.g. online forums and OSS repositories, and modern techniques such as data mining, offer unique opportunities to collect large amounts of qualitative data. However, these data are not custom elicited and will likely require more effort in selecting, filtering, and cleaning before the can be used for analysis.

A combination of traditional and modern approaches is recommended to achieve best results, where applicable. For example, an STGT study can use traditional data collection techniques such as interviews to elicit an initial set of rich and highly relevant concepts and (sub)categories during the basic stage. These can be used as seed terms to perform large-scale targeted data collection on public and open source data. Without the initial traditional data collection, it would be challenging to identify, select, and filter from the large numbers of and massive public and open datasets. Without modern data sources and techniques, it would be nearly impossible to manually scale the theoretical outputs, e.g. to be more robust and include a wider set of applicable contexts. A combination of both offers unprecedented relevance, robustness, and scale in theory development.

The open nature of modern data sources, where anyone can add any information with minimal authentication and scrutiny but with mass reach, poses new challenges for researchers. While traditional Glaserian GT accepts “all is data” (Glaser and Strauss, 1967), STGT nuances the original stance with the necessary caution required to assess modern data sources and accepts credible data. It also highlights the importance of ethical considerations, especially associated with modern data collection and usage (upcoming in section 7.3). That is, for STGT studies, all is data, with care.

Basic data analysis includes open coding and constant comparison. Coding is the process of representing textual raw data into condensed formats that best capture its essence and meaning. Using a socio-technical approach to coding enables the researcher to reach beyond analysing the social context and meaning. It ensures the technical aspects are neither ignored nor treated as a token or decorative element (e.g. like a black box), rather they are considered together with the social aspects to capture a more comprehensive socio-technical essence and meaning of the raw data.

Open coding refers to the coding applied in the early stages of the research study where the researcher remains open to any and all codes arising, ensuring a comprehensive coverage. Because of its open nature, open coding is likely to result is large amounts of code. Constant comparison is the process of constantly comparing derived codes within the same source and across sources to identify key patterns in the data. Constant comparison concretely manifests an inductive approach, leading from specific instances toward general patterns, within the study context. Constant comparison is applied at increasing levels of abstraction to raise the codes to the concept level, concepts to sub-category (where applicable) and category levels.

Application of a traditional sociological approach to GT data analysis in a software engineering research context is likely to identify social concepts and come up with rich theories comprising them. For example, applying open coding the traditional way can lead to the identification of concepts that capture how software development teams regularly engage in social activities of brainstorming, discussing, debating, and reconciling choices. However, software engineering is not simply a mixture of social and technical aspects studied in isolation. Socio-technical data analysis in STGT refers to analysing the complete socio-technical context. For example, applying STGT in the same context is likely to uncover and explain the full socio-technical nature of the same activities as brainstorming software requirements, discussing software architecture, debating low fidelity prototypes, and reconciling choices of diverse technologies and platforms. These concepts encapsulate more comprehensive socio-technical meanings and essences of the activities being observed and are more likely to be relevant to and benefit practitioners, a key aim of GT studies. Researchers applying a socio-technical approach to open coding are likely to apply high levels of socio-technical sensitivity, picking up on the nuances of different socio-technical practices involving seemingly similar social activities, e.g. debating low fidelity prototypes as opposed to debating design patterns. Where developers may be biased toward the prototypes they create themselves and therefore more defensive in their debating strategies in the former case and more open to logical argumentation when debating established design patterns in the latter case.

Interestingly, most quality GT studies in software engineering seem to ‘naturally’ apply a socio-technical adaptation to data analysis, although rarely acknowledged and hitherto undefined and unexplained, leading to the formulation of socio-technical grounded theories, e.g., theories of reconciling agile and software architecture (Waterman et al., 2015), becoming agile (Hoda and Noble, 2017), and variations in scrum practice (Masood et al., 2020b). More examples of applying a socio-technical approach to open coding can be found in (Hoda et al., 2011, 2012b; Hoda and Noble, 2017; Masood et al., 2020b, a; Shastri et al., 2021b, a).

Explicitly acknowledging and explaining the data’s socio-technical nature will enable aspiring researchers to conduct quality STGT data analysis. Additionally, while it was considered “very shortsighted to believe that computers are capable of gleaning the meaning embedded in qualitative data” (Becker, 1993), rapid technological advancements in natural language processing, sentiment analysis, and other AI based analysis techniques, promise further easing and augmenting of manual analysis, enabling unprecedented improvements and scaling of qualitative data analysis and theory development in the future.

Basic memoing is the process of documenting the researcher’s thoughts, ideas, and reflections on emerging concepts and (sub)categories and evidence-based conjectures on possible links between them. Memos can be in the form of written notes, sketches, annotated images or photographs. While voice and video recording can also be used, it is recommended these are transcribed for ease of further comparison and analysis in the advanced stage.

Memoing is an imperative procedure that distinguishes STGT from other qualitative research methods. It is the mechanism through which researcher reflection is systematically documented and used for theory development. Examples of memos can be found in (Hoda et al., 2012a) and (Masood et al., 2020a).

Where the initial research phenomenon or topic was broad (e.g. exploring agile project management (Hoda et al., 2012a)) and open coding was applied extensively on the broad level, the basic stage will likely lead to the emergence of a number of (sub)categories, some stronger than others, with indicative relationships emerging between them. In this case, where a some strong categories and initial relationships have emerged but a clear theoretical structure is not evident, the researcher can decide to proceed with an emergent mode of theory development which relies on the theoretical structure emerging progressively over time.

Data collection in the emergent mode is driven by theoretical sampling as described earlier, targeting data sources most likely to help fill theoretical gaps and strengthen key categories, and including refinements of the data collection protocols to progressively focus on the key categories.

Targeted data analysis involves targeted coding which is the same procedure as open coding but applied in a targeted manner, and constant comparison. Unlike open coding, targeted coding²²2Targeted coding is the same as selective coding in Glaserian GT but STGT avoids using the term selective coding since it refers to two very different procedures in Glaserian GT and Strauss-Corbinian GT. To avoid confusion, STGT uses the term targeted coding. limits analysis to only the most significant concepts and categories arising from the basic stage. However, some new categories can still emerge during targeted data analysis. Advanced memoing ensures the links between (sub)categories are strengthened.

Once theoretical saturation is achieved, the researcher can decide to present the theory as is, using the emergent structure. However, STGT recommends the researcher to perform theoretical structuring by (a) exploring and identifying the genre of theories that best fits, e.g. process, taxonomy, degree, strategies (Glaser, 1978), and (b) exploring if the emergent theoretical structure naturally maps to a pre-existing theoretical template, e.g. six C’s model, process template with defined stages and entry/exit indicators, and in case of a good fit, use the template to further structure their theory. Theoretical structuring³³3Theoretical structuring in essence is the same as theoretical coding in Glaserian GT but STGT avoids using that term since the procedure does not involve any actual coding, rather structuring or organising of the emergent theory. is an optional but highly recommended procedure.

The theory of becoming agile is an example of presenting a mature theory in its emergent structure, i.e. a bespoke process model (Hoda and Noble, 2017). The theory was classified under the genre of process theory since it explained the process of software teams becoming agile teams but no explicit pre-defined structure was adopted. On the other hand, in case of the theory of inadequate customer collaboration (Hoda et al., 2011), the pre-defined six C’s theoretical template was adopted at a late stage as a result of exploring Glaser’s theoretical coding families and discovering a good fit with the six C’s template (Glaser, 1978).

Where the initial research topic was relatively narrow and well defined (e.g. investigating self-assignment practice (Masood et al., 2020a)) and the data collection was focused on well defined facets of the topic, the basic stage can lead to the emergence of a set of (sub)categories and a relatively clear theoretical structure. In this case, it makes sense for the researcher to proceed with a structured mode of theory development.

Data collection in the structured mode is driven by theoretical sampling, collecting data to progressively fit, add, strengthen, and saturate the concepts and (sub)categories already present in the guiding theoretical structure. It includes refinements of the data collection protocols to progressively saturate existing categories and occasionally identify new concepts that appear within the theoretical structure.

Data analysis in the structured mode involves structured coding and constant comparison. Structured coding is similar to the original axial coding introduced in (Strauss and Corbin, 1990) in that it applies data analysis using a guiding theoretical structure and aims to identify and strengthen the relationships between the key categories. Advanced memoing ensures the links between (sub)categories are strengthened.

Once theoretical saturation is reached in the structured mode, the theory should comprise of dense and strongly supported categories and hypotheses and is considered both structured and mature. But, it may not be fully integrated. Theoretical integration is the procedure of ensuring the (sub)categories are fully integrated into the overall theoretical structure. Practically, this involves finalising the story-line of the theory, answering the question, what is this theory about? or what phenomenon is captured and explained by this theory? For example, in case of (Masood et al., 2020a), theoretical integration toward the later stages involved discussing what basic phenomenon the categories and relationships were best capturing. How agile teams make self-assignment work was decided as the overall story-line or phenomenon that the categories and relationships best helped explain.

Sometimes, integration can result in the emergence of an additional theoretical layer that serves to support and enhance the existing structure. For example, the theory of Scrum variations was structured around three categories, variations in Scrum roles, variations in Scrum practices, and variations in Scrum artefacts (Masood et al., 2020b). In the final stages of integrating the theory, a nuanced classification system emerged which served to further classify the variations across the three categories in terms of standard, necessary, and contextual variations, and clear deviations.

Standalone STGT studies refer to a full STGT method application in a standalone capacity and do not apply other research methods such as survey research to further validate the theory. Mixed data, both qualitative and quantitative data, can be used within standalone studies. Standalone applications result in what Storey et al. (2020) refer to as descriptive knowledge, with potentially limited prescriptive knowledge in the form of practical guidelines and recommendations (e.g. as seen in Masood et al. (2020a), Masood et al. (2020b), and Shastri et al. (2021b).)

Standalone STGT studies can vary in scale. They can take the form of small studies that address narrow research topics, e.g. impact of human values on software design decisions, mob programming techniques, or the use of coupling and cohesion in microservices design, the last derived from a Twitter post by Grady Booch. They are conducted over a relatively small duration of time, e.g. six to eighteen months, typically by a single researcher or a team of researchers. Small standalone STGT studies are likely to result in one mature theory, usually presented in a single publication, although preliminary and partial results can be reported along the way.

They can also take the form of large scale studies that address broad research topics, e.g. role of ethics in artificial intelligence, future of work in a post pandemic world, and digital equity for the elderly. A comprehensive treatment of such topics requires several years, typically by a team of researchers. A standalone STGT study conducted as a part of a PhD program, typically three to four years in length, is an example of a large-scale STGT study. These are likely to result in multiple theories, usually published as multiple articles. How the multiple theories ‘hang together’ is usually detailed in the thesis report and should be briefly explained in individual articles with cross-reference to related articles.

There is an increasing trend of performing mixed methods, combined studies, and R&D projects in SE. In such cases, a full STGT study serves as a robust mechanism to research the problem space, leading to the development of highly relevant and customised solutions. For example, a small scale STGT study can be conducted as a first step to establish theoretical foundations, explain the problem in depth, and/or devise taxonomies followed by wider validation and/or tools development. Outcomes in these cases are expected to be what Storey et al. (2020) refer to as descriptive knowledge from the STGT research part that then ideally feed into or lead to the development of solutions, e.g. tools, techniques, platforms, from the application of other methods or in the development component of the R&D project.