A same-store cohort is the group of units, such as shops or companies, present for the whole of a comparison period. Measuring only that group means growth reflects change in the units themselves, not units joining or leaving. Retailers use it for sales, and data series use it so that wider coverage does not look like growth.
How it is calculated
- Fix a window, for example 30 days.
- Keep the units covered since before the window began and still covered at its end. That is the cohort.
- Count the event over the cohort in the window.
- Count it over the same cohort in the window just before.
- Divide the count by the cohort size for a rate, and compare the two counts for growth.
An illustrative case: a cohort of 400 websites, of which 24 add a tracking tool in the window. The rate is 6 per 100 websites. The same 400 added 20 in the previous window, so growth is 24 minus 20, divided by 20, which is 0.2.
Why it matters
A company dataset grows every day as companies are added. A raw count of tool additions rises when coverage rises, even if no company behaves differently: add a tenth more websites and the count rises by about a tenth. The cohort removes that effect.
It also limits what a rate says. It describes units that continued through the window, not those that entered or left, so it is not a measure of the net change of the whole market.
In Fokals data
The Market Series dataset holds a count, the cohort, a rate, the previous window, growth and an index for each metric, dimension and window of 7, 30 or 180 days, written once per weekly as-of date. The cohort is the companies or websites covered since before the window began and still covered at its end, and the previous window is taken over the same cohort.
A first observation of a site sets a baseline, so a newly covered site adds nothing to an adoption count. Each row rests on at least 20 companies, or 20 postings for pay and share metrics. The index is 100 at the first as-of date of a series. The method is explained in same-store cohorts and the weekly series.
Related terms
- Survivorship bias: leaving out the units that dropped away.
- Point-in-time data: rows kept as they were written.
- Data coverage: which companies a dataset includes.
- Technology churn: tools dropped, a rate the series count per 100 websites.
Frequently asked questions
What is same-store sales growth?
Same-store sales growth is the change in sales at stores that were open in both the current and the earlier period. Leaving out stores that opened or closed separates how existing stores are trading from how fast a chain is adding or losing shops. Analysts of retailers and restaurants watch it for that reason.
Why use a same-store cohort for company data?
Because the coverage of a company dataset keeps growing, so raw counts measure the dataset as much as the market. Taking rates over the companies covered throughout the window, and comparing the previous window on that same group, keeps a change in the series from being a change in coverage.
What does the index in a market series mean?
The index is 100 at the first as-of date of a series and then shows the rate relative to that start, so an index of 120 means the rate is a fifth higher than at the start. It puts series with different units on one scale, and because rows are written once, it does not change when you read it later.