Not every analytics question is about what drives the outcome
Some variables are not drivers — they are constraints. Choosing the right method starts with choosing the right question.
The Problem
Most data science work asks: "what increases Y?" That leads to correlation, regression, feature importance, SHAP.
But some questions are different: what must already be true for high Y to even be possible? That is a different analytical problem entirely.
The Method
In one enterprise survey project, I used Necessary Condition Analysis (NCA) to identify variables that acted less like average drivers and more like minimum requirements.
A regression coefficient tells you a variable is associated with better outcomes on average. An NCA result tells you something else: below a certain threshold, the desired outcome simply does not appear — regardless of everything else.
Regression
Innovation has a coefficient of 0.3 on growth = innovation helps on average
NCA
No company with innovation below 2.4 achieved growth above 7 = innovation is a necessary floor, not just a booster
Key Takeaway
Design insight: 'Invest in X to boost performance' is not the same as 'guarantee X above threshold before anything else matters.' Some variables are not drivers — they are constraints.