⌕
← Methodology topics
Statistics

Median, percentiles and IQR

Why we publish median and percentiles instead of mean prices, and how to read the IQR for variance signaling.

Why not the mean

Luxury handbag pricing is heavy-tailed: a small share of listings are wildly mispriced (test listings, accidental zeros, fakes with reach prices). The arithmetic mean inherits all that noise. The median is the price at which half of observations are higher and half are lower — a single auction at $99k does not move it. We treat the median as the canonical price and report all other statistics relative to it.

Percentiles P10, P25, P75, P90

We publish four percentiles around the median. P25 (first quartile) is the price below which 25% of observations fall; P75 (third quartile) is the price below which 75% fall. P10 and P90 capture the tails. Together they describe the shape of the distribution without assuming normality.

Interquartile range (IQR)

IQR = P75 − P25. It is the width of the middle 50% of the distribution. A narrow IQR indicates tight market consensus on price; a wide IQR signals condition-driven variance, brand-mix variance, or low liquidity. We use 1.5×IQR fences to remove outliers before computing the median (Tukey method).

Sample-size minimums

We require at least 30 observations in a 90-day window for medium-confidence aggregations and 100 observations for high-confidence labels. Below 30 we still compute the median but label confidence as low and dim the UI.

Other topics