Research method and coverage

Where the prices come from

Every study uses DeserTracker's own records of prices observed on public UAE online stores. We record a price when the price or the stock state of a tracked product changes, not on a fixed schedule.

Studies name no store. No study uses data from before March 2026.

How we pool stores

Studies pool prices observed on public UAE online stores and name none of them.

In a share, a median or another weighted number, except a price gap between stores (see Comparisons), each product counts once, a store joins only with enough products for that number, and no single store supplies more than half of the weight.

Counts, such as the number of cuts or of products, are plain totals over the stores in the study. One store can supply most of a count.

A store joins a study only when its collection coverage passes the check below and it has enough products for the headline numbers. Each number needs at least two stores that contribute to it. A store with a count of zero does not contribute to that count, except in a table row that other stores share, where zero is a true value. In the same way, a share or a rate, such as the share of products with a cut or the number of cuts per 100 product-days, is published only when at least two stores had a counted case. A store without a case still counts towards the value of a published share or rate.

How we weight time

A price counts for the time it was in effect. Between two recorded changes the product holds one price; we call that interval a spell. Medians and shares weight each price by the hours of its spell.

Time out of stock is not price time. An out-of-stock record ends the current spell, and the next in-stock price starts a new one.

Censoring

A spell that is still running at the end of the data window is unfinished. Unfinished spells leave the denominator of any duration statistic, because we do not know how long they last.

A cut that has not had its full follow-up time by the end of the window is not eligible for the return statistic of that horizon.

Coverage

Coverage is checked per store and day, not per product. A day that is not covered is left out: no spell runs across it.

Before the daily check counter started, we infer collection from the daily count of recorded price and stock changes. A day counts only when that count reaches at least 50% of the median of the preceding 28 days.

From the counter start, a day counts only when successful price checks reach at least 25% of the median of the preceding 28 days.

Known collection gaps count as uncovered with any count. A store whose uncovered share of the window is above 20% does not join the study.

Each study lists its collection gaps by date range and reason only.

A price recorded before the window carries into it for at most 30 days, and only when every day between the record and the window start is covered.

What counts as a cut and a return

A cut is a price change that is at least 1% and at least 1 AED below the previous price. Values exactly on a threshold count.

A full return means that within the horizon the price is back at or above the price before the cut. A partial return means that it recovers at least 50% of the cut.

A store needs at least 100 products to join a study's headline numbers, and at least 100 products for a share, a median or another weighted number, a table cell included, to join that number. Many observations of fewer products do not count. When fewer than two stores qualify for a number, the study gives no value for it. When that happens to a headline number, the study reports too few stores. In a table, such a cell shows "withheld", and a row that has no other weighted cell is left out.

What we exclude

Third-party backfill records, synthetic first points, quarantined records, prices older than the carry limit, and stores that DeserTracker does not track.

Comparisons

Results are pooled across stores. We publish no number for a single store.

A price gap between stores uses only verified matched products in shared categories. Each matched pair counts once. Pairs are pooled by the pair of stores they compare: a pair of stores joins a number only with at least 100 matched pairs for it, and no single pair of stores supplies more than half of the weight, so one store that is in every pair can supply most of it. A pair counts towards "the cheaper store stayed cheaper" only with at least two sampled moments. A share is published only when at least two pairs of stores had a counted pair.

Wording

We write "lowest observed since <date>" for the lowest price in our records. We never claim an all-time low, and we make no claim about the intent of a store.

Method version

This page describes definition set m3. Each study names the method version it was computed with.

Studies published under m2 show the same numbers; m3 only adds the rule that a share or a rate needs cases from at least two stores.

Corrections

Found an error? Email [email protected]. A corrected study shows the date of its update.

Studies that use this method

• How long price cuts last at tracked UAE retailersWe observed 1,582,803 price cuts; each cut counts once.
• How much of the time prices sit at their lowest observed level at tracked UAE retailersWe followed 569,522 products with an in-stock recorded price at the window start.
• How prices moved in the weeks before the last week of June at tracked UAE retailersThe comparison includes 515,047 products.
• Does waiting a week or a month get a lower price at tracked UAE retailers?After a week, the observed price was lower in 16.2% of eligible comparisons.