Theory Testing vs. Theory Building: When Each Is Appropriate

There is a default research design that dominates IS scholarship: take an existing theory, add one or two constructs that seem relevant to a contemporary phenomenon, develop hypotheses, collect a survey, run structural equation modeling, and report results. This is theory testing. It is a legitimate and valuable form of research. But IS has arguably over-indexed on it for thirty years, and the incremental contribution of each new study has become very small.

The distinction between theory testing and theory building matters because they are different epistemic activities. Theory testing takes theoretical claims that already exist and asks whether they hold in new samples, new contexts, or under new conditions. Theory building discovers new constructs, new relationships, or new mechanisms that were not previously captured in any formal theory. Both are necessary for a field to accumulate knowledge. The problem is that they are not equally valued or equally common in IS research.

Eisenhardt (1989) is the classic reference for building theory from case studies, and it is confirmed as local StudyHub evidence in IS research methods. The paper describes a systematic process: select cases purposively, do intensive within-case analysis before cross-case comparison, look for patterns and contradictions, and stop adding cases when you reach theoretical saturation, meaning new cases no longer add new theoretical insights. The output is a set of propositions or constructs that have emerged from the data rather than from prior theory. Eisenhardt (1989) is explicit that the process should stay in "controlled opportunism," letting the data guide theoretical development while maintaining enough rigor to produce defensible results rather than arbitrary storytelling.

Yin (2018) provides the systematic logic for case study research more broadly, and the study-hub confirms this as local evidence as well. Yin's point is that a case study is an empirical inquiry into a phenomenon in its real-world context, especially when the boundary between the phenomenon and the context is not clear. This is exactly the condition under which theory building is most valuable: when you do not yet know what the relevant variables are, when the mechanisms are unclear, when the phenomenon is new enough that borrowing from existing theory might just force the new thing into old categories.

Gregor (2006) provided a taxonomy of IS theories that helps clarify what kind of knowledge different research activities produce. The five types range from analysis and description (Type I) through explanation (Type II) and prediction (Type III) to explanation and prediction together (Type IV) and design and action (Type V). Theory building tends to produce Type I and Type II contributions first, because you have to understand and describe a phenomenon before you can explain it. Theory testing is more naturally connected to Type III and IV contributions, where the causal relationships are already specified and you are checking whether they hold.

The TAM story is instructive here. Fred Davis (1989) built the Technology Acceptance Model through a sequence of qualitative and quantitative work that was genuinely inductive at key points. He was asking what determines whether people will use a new computer system, at a time when that was not a well-understood question. He developed perceived usefulness and perceived ease of use as constructs through item generation and refinement processes that were grounded in what actually drove people's assessments of the systems they used. That is theory building. Not pure grounded theory in the Glaserian sense, but genuinely generative work.

What the field did after 1989 is a different thing. Hundreds of papers extended TAM by adding moderators, antecedents, and boundary conditions. Social influence was added. Trust was added. Hedonic motivation was added. Cultural dimensions were added. Computer anxiety was added. Each individual extension followed the testing logic: TAM already exists, here is a new construct that might matter, here is a hypothesis, here is a survey, here is the result. The contribution of any one of these papers is modest. The problem is that after enough extensions, it is not clear that the cumulative enterprise is building toward a more complete theory of technology acceptance or just expanding a list.

This is not an abstract criticism. It has practical implications for how IS research is evaluated and what gets published. A paper that tests TAM with one new moderator in a new context (medical professionals using electronic health records in rural Thailand) can find a home in a decent journal if the sample is large enough and the statistics are clean. A paper that says "TAM cannot explain this phenomenon, here is a new construct you have never seen before, here is where it came from in the data" faces more skepticism, because it is harder to evaluate and it requires the reader to accept something new.

Theory building requires a tolerance for ambiguity. You spend time in the field, or in archival data, or in qualitative interviews, without knowing yet what your theoretical contribution will be. That uncertainty is productive but uncomfortable. The case study and grounded theory traditions, which I have written about separately, are designed for this kind of work. They are methods suited to situations where you do not yet know what you are looking for in enough detail to test it.

The practical advice I give myself when thinking about a research question is to ask whether existing theory can already explain the phenomenon or whether something genuinely new seems to be happening. If users are behaving in ways that TAM or UTAUT or the DeLone and McLean success model predict, then testing those models in the new context is an appropriate contribution, even if it is modest. If users are doing something that none of the existing models anticipate, that is a signal that theory building might be warranted. The anomaly is the invitation.

What bothers me about the field's current balance is not that theory testing exists. It is that the field sometimes treats theory testing as the default and theory building as the unusual case that requires extra justification. It should be the opposite. If you already know what you are going to find before you collect the data, ask yourself whether that is actually a research contribution or a confirmation exercise. Sometimes confirmation has value. But it should not be the dominant mode of a field that is supposed to be generating new knowledge.