Trifurcation discards most of the scale
Aliases: NPS three-bin scoring · promoters and detractors · coarsened recommend scale
What it is
NPS scoring first cuts 0–10 into three bins: detractors 0–6, passives 7–8, promoters 9–10, then subtracts the detractor percentage from the promoter percentage. 0 and 6 count as the same class, 9 and 10 as the same class, and 7–8 never enter the numerator. Eleven ordered judgments collapse into one signed percentage-point difference. Means, medians, and counts at each tick—structure still present in the raw answers—are thrown away at the first calculation step. The same NPS can be produced by entirely different underlying distributions.
Why it happens
On an ordered scale, adjacent numbers usually mean closer attitudes. Trifurcation cuts that distance structure: the one-step move from 6 to 7 crosses a class and weighs heavily; a six-step move from 0 to 6 inside the bin weighs zero. Teams are therefore rewarded for turning a 6 into a 7, or an 8 into a 9, rather than a 0 into a 4. Passives only pull the score toward zero by enlarging the denominator; whether they sit at 7 or 8 disappears as well. Subtraction also lets different mixes print the same result: promoters and detractors rising in the same proportion can leave NPS almost still. Information is discarded twice, once in the cut and once in the subtraction.
Studying it
Report the raw 0–10 distribution and the mean alongside NPS, and test whether they point at the same change. Simulations or historical files can hold NPS fixed while moving quality (lifting every 0 to 6, or dropping every 10 to 9) to see whether decision makers still issue the same judgment. A transition matrix across bins (who moved from which tick to which) describes what happened better than the total. If the research question is attitude intensity, analyze the raw scores and treat trifurcation as a derived display, not the other way around.
Where it stops holding
When an organizational protocol already locks NPS as the trending figure, the number can still be reported for continuity; the analysis layer should still keep the raw distribution. In tiny samples, one or two extreme zeros become large after trifurcation, more volatile than a mean. Applying the 9–10 cut to a 1–5 grid is a different ruler and must not be mixed with standard trifurcation. What the bins discard is intensity inside the scale, not cross-cultural response styles, and not industry benchmark values.
Applying it
- Default the dashboard to a 0–10 histogram, with NPS as a derived number beside it, never full-screen alone.
- Trigger alerts on the mean and on bin counts as well. If NPS moves while the underlying distribution barely does, label it a “binning artifact.”
- Do not write improvement goals as “turn 6s into 7s.” Treat deep dissatisfaction and marginal hesitation at their actual positions on the raw scale.
- External trends must carry the distribution or at least bin shares. An NPS point with no distribution is not a sole basis for a decision.