Mohr warned us in 1982 that variance and process theories answer different questions and need different evidence. Most IS research still evaluates process stories with variance criteria. Here is why that is a mistake.
I kept reading the same sentence in Mohr (1982) and realizing I had been making the exact mistake he warned about. Variance theory needs necessary and sufficient conditions. Process theory needs necessary but not sufficient conditions, and the order of events matters. Two different logics, two different evaluation criteria, and most of us in information systems treat them as interchangeable.
This is not a minor methodological point. It is the difference between explaining who is likely to churn and explaining why an enterprise system collapses.
Variance theory operates in the "if X, then Y" register. More education leads to more income. Higher perceived usefulness leads to more system use. The conditions are necessary and sufficient: given enough X, Y follows, and the strength of the relationship is measured by how much variance in Y is explained by X. You run a regression, you report an R-squared, and you have a clean story. This logic works for prediction. A churn model at a telecom company can tell you which customers are likely to leave next month based on usage patterns, billing history, and support interactions. The model does not need to know why each person leaves. It needs correlations strong enough to act on.
Process theory operates in a different register. It says that certain necessary conditions must be present, but no set of conditions is sufficient to guarantee the outcome. Instead, the outcome emerges from a temporal sequence of events. A divorce does not happen because of a dose of dissatisfaction. It happens because a sequence of events, each necessary but none sufficient on its own, led to a rupture. The early argument that established a pattern of avoidance. The refusal to repair. The slow redefinition of the relationship. Each step was necessary, but none alone was sufficient. The order matters: rearrange the sequence and you get a different outcome.
Mohr made this distinction in 1982, and it still catches researchers the wrong way. The trap is evaluating a process study with variance criteria. You read a detailed process tracing of how an ERP implementation failed, and you ask how much variance it explains. That is the wrong question, and it is the wrong unit of evaluation. Process theory is judged by the fidelity of its narrative to the actual sequence, by the coherence of the story, by whether the identified necessary conditions were present and in the right order. It is not judged by R-squared.
Markus and Robey (1988) placed this distinction inside the broader causal structure framework that every IS researcher should carry into an exam or a review. They identified three dimensions along which any causal claim can be positioned. Causal agency asks who or what drives the outcome. The technological imperative says technology determines outcomes on its own. The organizational imperative says human choices determine outcomes. The emergent perspective says outcomes arise from the ongoing interaction of technology, people, and context. As I wrote elsewhere, the emergent view is almost always the correct IS answer, which means most real phenomena we study are not pure variance or pure process but a combination that requires care in specifying which logic you are using and where.
Logical structure is the second dimension. This is where Mohr's variance versus process distinction lives. Are you explaining an outcome through the strength and direction of relationships among variables, or are you explaining it through a sequence of events where order matters? The third dimension is level of analysis: individual, group, organization, or industry. A theory that is valid at the organizational level does not automatically transfer to the individual level, and vice versa. Each dimension constrains what counts as evidence, and mixing them without acknowledgment is one of the most common exam mistakes.
The reason this matters beyond the exam room is that most organizations evaluate everything with variance logic. Startup post-mortems read like failed dose-response experiments. "We had the right team, the right product, the right market timing, so why did we fail?" The answer is that none of those conditions was sufficient. They were necessary, but their temporal configuration, their sequence, mattered. The team that was right for the seed stage may have been wrong for the growth stage. The product that fit the market in Q1 may have missed it by Q3 because the market shifted. The timing that looked perfect in the business plan collapsed because a key partnership fell through in week seven, not because macro conditions changed.
Theranos is a useful case not because of the fraud but because the fraud followed a sequence. Elizabeth Holmes did not set out to build a fraudulent company. She set out to build a blood-testing revolution, encountered the reality that the technology could not do what she promised, and then made a series of choices, each one understandable in isolation, each one necessary for the eventual collapse, but none sufficient by itself. The decision to use commercial analyzers for validation while claiming proprietary results. The decision to intimidate employees who raised concerns. The decision to mislead investors about accuracy. Each step built on the last, and the order mattered. You cannot predict Theranos from a regression of startup characteristics. You can only explain it by tracing the sequence.
This is also why predictive analytics in enterprise software keeps disappointing organizations as an explanatory tool. A churn model can tell you who will leave. It cannot tell you why. An adoption model can predict how many users will log in next quarter. It cannot tell you why the system became irrelevant in one department but essential in another. The predictive model does variance logic well. The organizational failure follows process logic, and evaluating it with variance criteria produces the same category error Mohr identified in 1982.
I think most IS researchers evaluate process studies with variance criteria because variance logic is what our training rewards. Journals want effect sizes, variance explained, falsifiable hypotheses. A process tracing paper arrives at the review table, and the first question is often "What is the dependent variable?" The answer is that there may not be a single dependent variable in the variance sense. The outcome is a process, not a point, and the explanation is the sequence, not the coefficient. The paradigm you work in shapes what you count as evidence, and positivist training has a hard time crediting narrative coherence as a form of rigor.
The written answer in my study materials puts it plainly: process theory needs necessary but not sufficient conditions where the outcome comes from a temporal sequence, and it is judged by fidelity to that sequence and narrative coherence. A process study judged by variance criteria looks weak. A variance study judged by process criteria looks ahistorical. The two logics cannot be combined into one research design without breaking one of them. The IT artifact question only makes this harder, because when you start theorizing technology as something that actually does work in a social system, as Sarker et al. (2019) argue you must, the causal story almost always becomes process-like and emergent rather than variance-like and deterministic.
About the author
Share
More notes
Related notes