Immanuel Kant
Thoughts without content are empty; intuitions without concepts are blind.
A simple question without a simple answer
When we want to extract information, knowledge and finally insight from a phenomenon, one basic question is always in the way: should we first build a theory of “what” and “how” things should be measured, and then go gather data? Or gather data first and let the data itself tell us what the theory is?
These two paths are two real traditions in the history of data analysis. The labels “European model” and “American model” that are sometimes pinned on them are not an official, agreed classification, though; they are more an instructive generalisation, and below I say where it holds and where it doesn’t.
Path one: theory before data
In the classical tradition of philosophy of science, much of which formed in Europe, a hypothesis or theory is built first (for instance how some concept can be measured), then data that can confirm or reject it is gathered, and the analysis serves to test that theory. This is the hypothetico-deductive method. Karl Popper, the Austrian-British philosopher of science, made it famous with the principle of falsifiability: a theory is scientific when you can specify what data would refute it. (The method itself is older than Popper; Popper gave it its well-known form.)
Path two: data before theory
Against it stands a tradition that says too many theoretical assumptions can blind us to the raw data: first gather data, immerse yourself in it, and let theory emerge from it.
The best-documented example of this approach is grounded theory, introduced in 1967 in America by Barney Glaser and Anselm Strauss. Strauss came from the Chicago school and American pragmatism, and Glaser from Columbia. Their claim was that instead of going into the field with a ready-made theory to test, you go and let categories and theory take shape gradually from coding field data.
Decades later, in the Big Data era, the same idea returned with a louder voice. Chris Anderson, then editor of Wired, wrote in a controversial 2008 article, “The End of Theory,” that with enough data (Google’s, for example) there is no longer any need to form hypotheses and correlation is enough. That claim drew serious criticism: spurious correlation and not understanding why, which are exactly where plenty of data without theory ends up.
| Comparison | Theory first (deductive) | Data first (inductive) |
|---|---|---|
| Starting point | Hypothesis or theoretical framework | Raw data |
| Role of analysis | Testing the theory | Discovering patterns |
| Known roots | European philosophy of science (Popper) | American pragmatism (Glaser and Strauss) and later Big Data culture (Anderson) |
| Main risk | Blindness to unexpected patterns | Spurious correlation without understanding why |
These roots are real, but we shouldn’t turn them into an absolute geographic rule (“all of Europe thinks this way, all of America that way”). That is oversimplification, and data analysis should be as alert to the traps of its own simplifications as it is to patterns.
What does it mean for a systems analyst?
This is the same tension we saw in the piece on Q-Analysis: first build a theoretical framework (simplex, face) and pour the system into that mould, or let the structure show itself from the real data of relations? I think the best systems analysts are neither purely theory-driven nor purely data-driven; they go back and forth between the two: make a light first theory, test it against data, correct it and go back to the data.
theory-drivendata-drivenmethodologygrounded-theoryfalsifiabilitybig-data