The Problem

When I first joined the leisure industry around ten years ago, I struggled to make sense of the length of stay (LOS) figures in industry reports. My CRM would report huge numbers sometimes over 3 years, as an "average" length of stay. These figures didn't match what I was actually seeing on the gym floor and every year they got slightly better.

If average length of stay really was that long, how could any operator have a retention problem?

The penny eventually dropped, the length of stay was being calculated using the mean. The mean is a perfectly good statistic for a fixed data set. For example, what was our average sale value yesterday?. However gym memberships are not a fixed data set. Members join and leave on every day of the year some joined yesterday, others joined five years ago. Using the mean on a moving, ongoing population like this produces numbers that look impressive but are structurally misleading.

I couldn't understand how anyone was running accurate cohort analysis or year-on-year comparisons without errors. I was sure the median was the right measure, but assumed as a newcomer to the industry that there must be a good reason everyone was using the mean instead.

It turned out there wasn't. When I met Dr Paul Bedford (The Retention Guru), I found someone who had reached exactly the same conclusion and had the same frustration with the mean. Hearing Dr Melvyn Hillsdon speak confirmed it further, there were credible voices in the industry already making the case for better data.

Mean vs Median: The Two Ways of Calculating Length of Stay

The Mean (what most gyms use)

The mean is calculated by adding up the total number of member-months across all members, then dividing by the number of members.

 That looks like a simple, reasonable calculation. The problem is what happens when you do comparisons between clubs, cohorts, or time periods.

The Median (what I believe should be used)

The median takes length of stay as the point at which 50% of a given group of members has left the club. Instead of averaging every member's tenure together, it asks a single question: how long did it take for half of this cohort to leave?

Example: Two Identical Clubs

Take two clubs that are, in every respect, performing identically. The only difference is that Club 1 opened one year ago and Club 2 opened two years ago.

Year One — Club 1 vs Club 2 (identical joiners and leavers)

 Club 2 has the exact same joiners and the exact same leavers in its first year, so it produces the identical mean of 9.6 months. So far, so consistent.

Year Two — Club 1 only (now with a second annual cohort layered in)

Club 1 is now in its second year of trading. A new cohort of 100 joins in January, and critically the mean calculation now also has to carry the 30 long-standing members from Year One who are still with the club and have now reached 24 months' tenure.

 This isn't a quirk of this specific example it's structural. It means an operator could easily conclude that one club, or one membership type, is genuinely outperforming another, when in reality performance is identical and the only difference is how long that club or membership type has existed. A membership type you've been selling for longer will always show a longer mean length of stay than a newer one  even if the newer membership actually retains members better, simply because it has more long-tenured survivors to inflate the average.

The Same Two Clubs, Calculated Using the Median

Using the median, we ask instead, at what point had 50% of each cohort left? Applying that to the identical scenario above:

Year One — Club 1 vs Club 2 (identical joiners and leavers)

 Club 2's Year One figures produce exactly the same result: a median of 9 months.

Year Two — Club 1 only (now with a second annual cohort layered in)

Why the Median Is the Right Measure

The core benefit is validity. The median only draws conclusions from members for the amount of time they could actually have been a member. If someone joined two months ago, they can only ever contribute to your 1-month and 2-month calculations, the median doesn't make silent assumptions about their future behaviour, unlike the mean, which effectively bakes in an assumption that current members will keep behaving the way they have so far.

This has two major practical benefits for operators:

• Fair comparisons
  • You can compare clubs, sites, or membership types on a like-for-like basis, regardless of how long each one has existed.
• Month-by-month visibility
  • because the median is built from cohort survival data, you can track exactly how retention is trending month by month, rather than relying on a single lagging headline number.

If the industry keeps reporting length of stay using the mean, operators will keep drawing the wrong conclusions from their own data. Rewarding the wrong clubs, promoting the wrong membership types and misjudging their real retention performance. Moving to the median isn't a statistical nicety, it's the difference between decisions based on real performance and decisions based on an artefact of how long something has existed.

Gavin Davies

COO 

guruPaul