How much does a commercial-only pairing rate vary from one vendor group to another?

The headline figure from this study, that 38.1% of vendor pairs cluster together only under commercial-intent framing, is a pooled average across every pair-by-group observation in the dataset. Pooled averages can mask a lot, especially when the underlying groups are heterogeneous. This study also reported the spread of that same rate measured separately within each group of vendors, which lets us check whether the average describes a typical group or is instead a blend of very different groups.

The answer is that it is a blend. Across the 28 groups where this was measured, the share of pairs clustering on commercial intent only ranged from 0% at the low end to 100% at the high end. The median group sat at 21.1%, noticeably below the pooled average of 38.1%. The mirror-image pattern shows up for pairs that cluster on both intents: across the same 28 groups, the both-intents rate ranged from 0% to 100%, with a median of 78.9%, above the pooled average of 61.9%.

The fact that some groups sit at the floor of zero and others at the ceiling of the full group tells us this is not a phenomenon with a single consistent rate that noise scatters around. It is a phenomenon that behaves in a qualitatively different way depending on which group of vendors you're looking at. A group at the zero end for commercial-only clustering is a group where every pair that clusters at all clusters on both intents, meaning intent framing simply does not move the pairing pattern for that category. A group at the full-share end is the opposite: essentially none of the observed pairings survive outside commercial framing.

Why would the median sit below the pooled average for the commercial-only class, and above it for the both-intents class? That pattern is consistent with the segmentation finding elsewhere in this study, where tightly grouped vendor sets showed near-universal both-intents clustering while loosely grouped sets split close to evenly. If the groups with more total pairs, meaning more vendors or more densely interconnected competitive sets, tend to skew toward the loosely grouped, more commercial-sensitive end, they would pull the pooled average toward a higher commercial-only share than the typical group actually shows. This study's group-level averages are consistent with that mechanism but do not, on their own, prove it, since group size and grouping tightness were not cross-tabulated in what was reported here.

The practical consequence is that a company should not treat the pooled figures, 61.9% and 38.1%, as a prediction for its own category. Those numbers describe the dataset as a whole, built from 43 groups averaging 32.33 units each, with no single group dominating the total since the largest held only 8.3% of it. Given the wide spread documented here, a company's own category could plausibly sit anywhere from a group where intent framing barely matters to one where it matters enormously, and the only way to know which is to check the specific pairings in that specific category, not to extrapolate from the study's average.