Methodology

01Data universe

The Index is computed from two of Fridge.com’s own data stores: the live product catalog (active refrigeration appliances) and a continuously collected NEW-condition price history from a major online retailer. No third-party index, survey, or estimate is used.

02Matched-model method

The Index tracks price change on a fixed set of individual models rather than an average of whatever is listed each month, so that a change in catalog composition (new models added, old ones dropped) cannot masquerade as a price change. For each pair of consecutive months, only models with a valid price in both months contribute. A model is identified by its canonical product record, which collapses retailer duplicates and finish variants of the same appliance into one model.

03Price definition

A model’s monthly price is the median of its daily NEW-condition prices within that calendar month. When a model maps to more than one listing, the daily price is the lowest available that day. A model-month is only used when it has at least 5 daily observations; below that a monthly median is too sensitive to single points.

04Source hierarchy

For each model-month the price is taken, in order, from: (1) the parsed NEW-marketplace price history (the primary, dense source — a full record of every price change, carried forward to a daily series); (2) daily NEW-price snapshots, used only where the primary source is thin for a month, and as an independent cross-source check; (3) retailer price history, which would be a separately labelled series but is not available in this version (the retailer price-history table is absent from the current database). The Index is therefore NEW-condition only, from a single major online retailer. The source used is recorded for every datum.

05Index construction

Each category index is a chained matched-model (Jevons) index: for month t, the index equals the previous month’s index multiplied by the geometric mean of each matched model’s price ratio (price in t divided by price in t−1). The base month is set to 100. The published window is floored (deeper price history exists but is not published, to bound survivorship bias from building a matched-model index on the current catalog). A category present at the base month anchors there; a category first stocked later anchors at its own first month. If a category later hits a month with no matched pair, its chain stops rather than inventing a value.

06Categories & publication threshold

Primary categories follow the catalog’s product types (french-door, side-by-side, top-freezer, chest-freezer, wine-cooler, and so on). Rollup categories aggregate related types: all refrigerators, all freezers, all specialty refrigeration, and all commercial refrigeration. A category-month is marked publishable only when it has at least 30 matched models that month; smaller categories are still computed, flagged below the publication threshold, and disclosed rather than hidden.

07Gates & quality checks

Every exclusion is counted and logged in the release metadata. The gates below render directly from this release’s dataset metadata, so the documented thresholds always match the computed figures.

Price gate

$20.00 – $30,000.00Prices outside this range are treated as data errors and dropped.

Ratio gate

0.5× – 2×Model month-pair price ratios outside this range are dropped as artifacts.

Minimum daily observations

≥ 5 per model-monthBelow this a monthly median is too sensitive to single points; the model-month is not used.

Publication threshold

≥ 30 matched modelsA category-month is marked publishable only at or above this count; smaller categories are still computed and disclosed.

Imputation

noneMissing prices are never invented; a chain with no matched pair simply stops.

Seasonal adjustment

none (v1)Readings are raw month-over-month movements.

As a release quality check, model-month medians derived from the primary price history are compared against the independent daily snapshots on an overlapping window; where they disagree beyond tolerance the winning source is recorded and used consistently.

08Known limitations

• The Index is NEW-condition only in this version, from a single major online retailer; there is no separate multi-retailer price series (the retailer price-history table is absent from the current database).
• The matched-model set is survivorship-limited to models in the current active catalog that carry price history; the published window is floored to keep this bias bounded but not eliminated.
• History depth differs by model; a model contributes only to months in which it has enough daily observations, and the honest earliest month the Index can cover is stated with each release.
• All matched models are weighted equally (unweighted geometric mean); the Index measures list-price movement, not sales-weighted realized price.