The Denominator Problem: What Attrition Numbers Actually Mislead

A programme that starts with 40 referrals and ends the quarter with 12 regular attenders can be written up two ways. One version says “70% attrition, cause for concern.” The other says “12 people are now attending a peer support group weekly, up from zero.” Both are true. Neither is useful on its own, because neither tells you what the denominator actually counted.

This is the more common failure than dishonest reporting outright. Most programme write-ups are not lying about attrition. They are counting the wrong thing and calling it a dropout rate.

The three numbers that get collapsed into one

“Attrition” usually gets reported as a single percentage, but it is doing the work of at least three different measurements, each of which implies a different fix.

1. Referral-to-first-contact loss — people referred who never attend at all. This is a reach and access problem: transport, timing, the referral itself not landing, or the person deciding on reflection that the offer wasn’t right for them.
2. First-contact-to-regular-attendance loss — people who show up once or twice and then stop. This is usually a fit problem: wrong group, wrong format, wrong week to have joined.
3. Regular-attendance-to-exit loss — people who were coming reliably and then stopped, sometimes because the programme ended, sometimes because their circumstances changed, sometimes because they got what they needed and left on their own terms.

A qualitative meta-synthesis of social prescribing found that participants described benefit extending well beyond social contact to restored purpose and structured engagement — which means someone who leaves after finding what they needed is not the same failure as someone who never engaged in the first place. Collapsing all three losses into one “70% attrition” line treats a person who achieved their goal the same as a person the programme never reached.

What this means in practice: before you write a single attrition figure, decide which of the three losses you are measuring and say so in the report. “58% of referrals did not attend a first session” is a different, more actionable claim than “42% of attenders stopped coming after week four,” and a funder reading either one should know which they’re getting.

The reporting habits that make the number misleading

Habit one: using the referral count as the sole denominator throughout. If 40 people were referred and 12 are still attending in month three, some reports state “30% retention” as though the 40 were ever a stable cohort to retain from. But referral numbers include people who were never a realistic match for the offer — inappropriate referrals, duplicate entries, people who moved away. Report the referral-to-attendance figure separately from the once-attended-to-still-attending figure, with a different denominator for each.

Habit two: silence on why people left. A systematic review of social prescribing’s effect on loneliness found that all nine included studies reported positive impacts for those who engaged, but the review also flagged the field’s habit of not distinguishing types of exit. If your intake process asks people why they’re not coming back — even a two-line note from the link worker — that data belongs in the report, categorised, not folded into a single percentage. “Left because moved area” and “left because didn’t feel it was for them” are not the same finding.

Habit three: no comparison point. A 70% dropout figure means something different depending on what similar programmes see. A difference-in-differences evaluation of a national loneliness campaign is one of the rare controlled looks at this kind of intervention at scale, and it is worth citing precisely because it shows how unusual it is to have any comparator at all in this field. Most attrition figures are reported in isolation, with no benchmark, which lets a mediocre number look either alarming or fine depending on the reader’s mood.

Habit four: treating the pilot’s attrition rate as if it will hold at scale. Small, hand-picked pilot cohorts — often recruited by a link worker who already knows the participants — attrite differently from a cohort recruited by broad referral once a programme is established. A protocol for reviewing social prescribing’s effect on isolation in older adults notes that despite wide adoption, only one peer-reviewed randomised controlled trial exists in this specific area. That is not a reason to stop measuring. It is a reason to state, explicitly, that your attrition figure comes from a small and possibly unrepresentative early cohort, and that it may not hold once the programme scales.

What an honest attrition section actually contains

Element	What it answers	Common omission
Referral-to-first-contact rate	Is the offer reaching people at all?	Reported together with later-stage loss
First-contact-to-week-4 rate	Is the format right for who shows up?	No data on which week people stopped
Categorised reasons for exit	Which losses are fixable, which are not	No exit data collected at all
Comparator or baseline	Is this rate normal for this kind of programme?	No benchmark cited
Cohort size and characteristics	Does this generalise past the pilot?	Cohort described only as “participants”

The named failure: the clean-cohort report

Programme teams under pressure to show results sometimes report outcomes only for the completers — the people who attended enough sessions to have measurable change — while burying the attrition figure in a footnote, or omitting it. Call this the clean-cohort report. It is the single most common way an evaluation misleads without technically lying: every number in it is accurate, and the picture it paints is false, because the people who dropped out are simply absent from the analysis. A randomised trial of volunteering as a loneliness intervention among older adults in Hong Kong is instructive here precisely because it is one of the few designs in this literature with a control group, which makes selective reporting of only the engaged arm impossible to hide. Most programme evaluations don’t have that structural safeguard. The discipline has to be supplied by the write-up itself: report the full referred cohort, then break out what happened to it, rather than starting the outcomes section with the survivors.

What this does not solve

None of this fixes the underlying reach problem. A cleaner attrition breakdown tells a funder more precisely who a programme lost and at what stage, but it does not put the programme in front of the people who were never referred in the first place — the isolated adults with no social worker, no GP visit, no route into the system that generates a referral. Roughly a third of adults 45 and older report loneliness in national survey data, and the people running programmes rarely see that full population; they see the fraction of it that reached a referral point at all. Honest attrition reporting makes a programme’s real reach visible. It does not extend that reach.