Diamond price drivers

918 natural pear-shaped diamonds, collected on 29 March 2026 from four online retailers. Every stone sits in a deliberately narrow band: 0.90 to 0.99 carat, colour D to K, clarity IF to SI2. The filter was applied at collection rather than afterwards, so the set is a slice of the market rather than a sample of it.

Per-carat prices across the set are right-skewed, with a skewness of 1.06. The mean is $2,571 against a median of $2,227, so the average is pulled up by a thin tail of premium stones.

A single-variable linear regression of price per carat on clarity, a second on colour, then a combined model with both. Each regression was run twice: once on grade averages, and once on the 918 individual stones.

Separately, pivot tables comparing the four retailers across the band, and a benchmark of collected prices against the Rapaport price guide. Built in Excel and reported in Word.

On grade averages clarity gives the tighter fit, 0.909 against 0.839, and colour has the steeper step. Ask the same question of the 918 individual stones and the two collapse to within 0.003 of each other, 0.335 and 0.332.

So which grade matters more has no answer that survives both bases. What does survive is the combined model: together the two grades explain 63.3% of the variation in price per carat, roughly double what either explains alone.

The two numbers answer two different questions: how cleanly grade averages decline, and how much of one stone’s price you can predict from its grades.

The four retailers do not agree with each other. James Allen averaged $1,550 per carat across the band and With Clarity averaged $3,249, a gap of roughly $1,700 per carat for stones inside the same narrow grade range.

Benchmarked separately against the Rapaport price guide, retail traded between roughly 5% and 35% below the published list. That is a different comparison from the retailer spread above, against an external reference rather than between the retailers, and the two should not be read as one result.

Averaging inside each grade removes the variation between stones that share it. What is left is a tidy line through seven or eight points, and it fits well because most of the noise has already been taken out. An R² of 0.909 against seven averaged points answers a different question from 0.335 against 918 stones, with a different denominator underneath it.

The cut premium is confounded. Excellent-cut stones averaged $3,358 per carat against $2,332 for very good, a premium of about $1,026, but the excellent-cut stones in this sample also carried better colour and clarity. The premium cannot be attributed to cut on this data.

The confidence interval is not the one that was planned. The intent was a 95% interval across the band, and the band did not contain 30 comparable stones. It was narrowed to a single colour-clarity cell, H-SI2, giving a mean of $1,760 per carat with an interval of $1,684 to $1,837.

Everything here came from a single pull on one day. These are listing prices rather than transaction prices, and what a stone is advertised at is not what it sells for.

The course paired two analysts on the same dataset and graded on whether both independent analyses produced the same results. The dataset was worked separately, the reports were written separately, and the findings were agreed by consensus afterwards.

That is why the figures here appear in another analyst’s record too, and why they match. The matching is the assessment, not a coincidence.