Glossary

Every term used in the reports, in plain language.

What I measure

Fit set: The price events that survived every check and actually produced the published figure. A report is only as good as this number.
Price elasticity: How strongly rank reacts to a price change, as a ratio. An elasticity of 1.0 means a 10% price change moved rank by about 10%. Below 1 means the reaction was weaker than the price move, above 1 means stronger.
Price event: One real price change I could measure: at least 5% and at least €0.30, held for at least a week, so short-lived flickers do not count.
Price segment: I split each category into three price groups (budget, mid, premium) by taking every product’s typical price and cutting the list into thirds. The boundaries come out of the data, so "budget" means something different in each category.
Sales rank (BSR): Amazon's position number for a product inside a category. Lower is better: rank 500 sells more than rank 5,000. It is a position, not a number of units, so a rank that improves by 20% does not mean 20% more sales.

How I judge it

Competitor reaction: How often rivals changed their own price after somebody moved first, compared with how often they would have anyway. Twice as often means the market answers; about as often means the timing was coincidence.
Control group: The products in the same category that did not change their price during the same days. They show what would have happened anyway, so that seasons, category trends and Amazon's own changes are not credited to the price.
Difference-in-differences: The comparison itself: how much rank moved for the products that changed price, minus how much it moved for those that did not. What is left is the part the price change explains.
Entry corridor: The price range where a new product would face the fewest established competitors and the lowest review counts, while the neighbouring prices still show real demand.
Estimate: The single best answer from the data.
Insufficient data: The category did not produce enough clean events to publish a figure at all. I say so instead of publishing a weak one.
Not demonstrable: The range runs into the area where nothing happens. I measured and cannot say whether an effect exists. This is a finding, not a failure.
Pre-trend: A product that was already moving unusually before its price changed. Those events are dropped, because I cannot tell the price from whatever was already happening.
Range (confidence interval): How precisely I know that answer. I publish 90% ranges: the true value very probably sits inside. Narrow means the events agreed, wide means they scattered.
Reference id: The code at the foot of every report. It points to the exact analysis run behind those figures, so a number can always be traced back.
Reliable: The range stays clear of zero, so I am willing to state a direction.