Store churn, mix effects, and the number that is not growth
Any measured universe of stores changes over time. Stores join, stores leave, stores close, stores change hands. A total computed over a changing set of stores moves for reasons that have nothing to do with demand, and the direction of that distortion depends on which stores joined.
This is one of the few data problems where the fix is well understood and the failure to apply it is common.
What does the distortion look like?
A total that rises when the panel expands and falls when it contracts, independent of what shoppers did. A provider that added stores in a strong region will report growth that is a composition effect.
The same applies within a brand's own reporting: a distributor gaining accounts shows volume growth that is a distribution effect, not a velocity one.
What is the like-for-like basis?
A calculation restricted to stores present and reporting in both periods being compared. It removes the composition effect and answers the question people usually mean: are the same stores selling more.
It is not free of assumptions. Which stores qualify — how many weeks of data, what completeness threshold — is a choice that should be documented, because different thresholds give different answers.
Why do stores drop out?
Closure, a change of system, a period of non-reporting, or leaving the panel. Those are different things with different implications, and lumping them together as churn hides the distinction.
A store that stopped reporting is not a store that stopped trading, and treating a reporting gap as a closure understates the market.
What is a mix effect?
A change in the total caused by a change in the composition rather than in any component. If larger-basket stores join the panel, average basket rises without a single shopper changing behaviour.
Mix effects appear at every level: store mix, item mix, region mix, channel mix. Any average computed over a changing composition is vulnerable.
How do you separate mix from rate?
With a decomposition that holds composition constant and varies rate, then the reverse. The two components sum to the total change and each is interpretable.
It is standard arithmetic and it is the single most useful thing an analyst can do to a surprising average. It converts "the average went up" into "the average went up because the mix changed, and rates were flat."
What should be published alongside a total?
The store count, the like-for-like count, and the churn in the period. Those three make a total interpretable and their absence makes it unfalsifiable.
A provider reporting the store base alongside the figure is being transparent about the one thing that could invalidate it, which is a signal worth noticing when choosing between sources.
How does this connect to same-store sales?
Directly — same-store sales is the like-for-like basis applied to retail sales, and it exists precisely because total sales conflate growth with expansion. A total that grew because stores joined is not growth.
Frequently asked questions
How many weeks of data should qualify a store?
It is a threshold choice rather than a fact, and it should be stated. Stricter thresholds give cleaner comparisons over a smaller base; looser ones give larger bases with more noise.
Does churn bias a panel in a particular direction?
It can, if joining and leaving are related to performance. Stores that close were often struggling, so a panel that drops them quietly can be biased upward.
What about stores that change ownership?
A definitional question: the same premises with different management may behave differently. Document the rule rather than deciding case by case.
Can I correct for churn myself?
If you have store-level data, yes, by restricting the comparison. If you have only totals, you cannot, which is a reason to ask what level of data you are buying.
Does this affect share calculations?
Less, because both numerator and denominator come from the same panel, but a compositional change can still move shares if the joining stores skew toward particular brands.
Is a growing panel a bad thing?
No — a larger panel is generally better data. What is bad is reporting a total across a changing panel as though it were a market movement.
How often should this be checked?
Every period, and it should be part of the standard reporting rather than an investigation. A store-count line on the same page as the total costs nothing.
What is the quickest red flag?
A growth figure with no store count beside it. Industry context for the independent channel is collected by NACS, and any credible read on it states its store basis.
### How does churn interact with a like-for-like series over years?
The qualifying base shrinks as the comparison period lengthens, because fewer stores are present throughout. A three-year like-for-like series rests on a smaller and increasingly unrepresentative set of long-surviving stores.
That is survivorship at the store level, and it tends to flatter. Reporting the qualifying base alongside the figure is what lets a reader judge how much weight it carries.
### How should churn be reported over time?
As a small standing table: stores in, stores out, stores reporting, and the like-for-like base. Four numbers on the same page as the headline, every period.
That table is also an early warning. A rising churn rate is a signal about the panel's stability that will eventually show up in the figures, and seeing it early is worth more than explaining it later.
Does churn matter more in some categories?
It matters most where store-to-store variation is high, because the composition change then moves the total more. Categories with concentrated distribution are especially exposed, since one joining or leaving store can carry a large share.
Working with retail data? See NRS Insights data solutions or contact NRS Insights.