The say–do gap, from Samuelson to your checkout funnel
In three points
- The gap between stated and revealed preference has ninety years of documentation behind it.
- Every field that took it seriously got better; UX measurement still treats it as noise to resolve.
- RUCF's contribution is not the gap — it is making the gap's direction prescribe the treatment.
In 1934, the sociologist Richard LaPiere toured the United States with a Chinese couple, visiting over two hundred hotels and restaurants. They were refused service once. Six months later he wrote to every establishment asking whether they would accept Chinese guests: over 90% of those who replied said no. What people said they would do and what they actually did were not merely different — they pointed in opposite directions.
Four years later, the economist Paul Samuelson formalised the other half of the idea: if you want to know what people prefer, watch what they choose, not what they claim. Revealed preference became a foundation of modern economics precisely because stated preference kept failing audits against behaviour.
Ninety years of the same finding
The attitude–behaviour gap has since been replicated across social psychology, health research, environmental studies and market research — fields that learned, sometimes expensively, that asking people is a measurement of what they say, not of what they do. Survey methodology grew an entire sub-discipline around the distortions: social desirability, recall error, question framing, adaptation.
UX measurement, a field younger than all of the above, largely skipped the lesson. Our standard instruments — satisfaction scores, effort scores, NPS — sample stated experience, and when behavioural data disagrees, standard practice is to “triangulate”: blend the readings until the disagreement disappears. Ninety years of evidence says the disagreement was the finding, and we average it away as noise.
From caveat to axis
RUCF's contribution is not discovering the gap — nothing on this site claims that. The contribution is structural: taking the say–do gap from a caveat in the appendix to the primary diagnostic axis of the model.
Measure friction only from behaviour. Measure perception only from report. Never let the two contaminate each other during collection. Then read the disagreement per surface, because its direction prescribes the treatment: perception better than behaviour means users have stopped noticing a real cost — subtract the cost. Perception worse than behaviour means the story is broken, not the flow — fix what the product says. The four resulting states, with a distinct playbook each, are the say–do matrix.
Your checkout funnel, 1938
Which brings us to the modern version of LaPiere's letters. Your post-purchase survey says checkout is fine: 4.2. Your funnel says 38% of started checkouts die before payment. You are holding both halves of a ninety-year-old experiment, and the question is only whether you will do what the establishments' letters did — report a clean answer that behaviour contradicts — or what the economists eventually did: believe the behaviour, and treat the stated answer as data about beliefs rather than about the product.
Both readings are valuable. They are just measurements of two different things, and the distance between them is the most informative number you own.
Signals this affects
The behaviour/perception pairing on all twelve signals — it is the reason each one is asked twice.
Related: Your users are not telling you what your product costs them · Stop reporting average UX