Fig 1.
Left: Distribution of top words for Planners & Organizers category for consumers. Right: Distribution of top words for Planners & Organizers category for vendors. This example illustrates the simplest application of Predictions 1 and 2. For illustration, we consider just the leading term. On each side of the market, “planner” was the lead term in 2012. And on each side of the market, this lead term was further strengthened in 2013. In this manner, the distribution of words on each side of the market became sharper, and the entropy of the word distribution on each side of the market decreased correspondingly. Underlined are words that are in the top-20 list for both consumers and vendors.
Fig 2.
Left: Distribution of top words for Handlebar Grips for consumers. Right: Distribution of top words among vendors. Vendors in 2012 used “grips” more than “tape”, while for consumers it was the opposite. Consumers learned from vendors and by 2013 they, too, showed a strong preference for “grips”. At the same time, vendors somewhat muted their strong preference for “grips” and began using “tape” in accommodation of consumers’ 2012 word usage. In the end, because consumers made a strong accommodation and vendors made a slight one, both sides ended up favoring “grips”, and thenceforth we would expect uncomplicated mutual convergence around “grips”. Underlined are words that are in the top-20 list for both consumers and vendors.
Fig 3.
Left: Consumer-size word domain distribution sizes per category percentile. Right: Vendor-size word domain distribution sizes per category percentile. We sorted categories according to the consumer-side word domain distribution sizes and reported on top 10%, top 50%, and all 100% of 11,838 categories.
Table 1.
Data descriptives.
Table 2.
Results for Prediction 1: Entropy changes, relative to uniform.
Table 3.
Results for Prediction 2: Changes in JS divergence in the two sides’ word distributions.
Fig 4.
Further analysis of Prediction 2: Percent reduction in JS divergence, for largest 1,184 objects.
Fig 5.
Further analysis of Predictions 1-2: The pattern of entropy / JS divergence increase/decrease between Vendors (V) and Consumers (C), for all objects.
Table 4.
Results for Prediction 3 regressions.
Fig 6.
Illustration of word distribution flattening before convergence.
Consumers lower the strength of their initial preference for A, while vendors lower the strength of their initial preference for B, i.e. increase their acceptance of A; this makes both sides closer to a 50-50 uniform split than they began, as they try to accommodate one another.
Fig 7.
Left: Distribution of top words for Prayer Beads category for consumers. Right: Distribution of top words for Prayer Beads category for vendors. Vendors in 2012 used “mala” more than “beads”, while consumers used both equally. Consumers learned from vendors and by 2013 they showed a preference for “mala”. This change represents a sharpening of their word distribution, compared with their initial equal division between “mala” and “beads”. The result was a decrease in entropy on the consumer side. On the vendor side, the opposite effect was observed. In 2012, vendors had had a clear preference for “mala” over “beads”. By 2013, learning from consumers, vendors had muted their relative preference for “mala” by increasing their use of “beads”. The result of this flattening of the distribution was an increase in entropy on the vendor side.
Table 5.
Each cell contains number of categories meeting those conditions.