Members who attend your annual meeting renew at a higher rate than members who do not. For
association member retention, the question is how much of that gap the event produced, and the
honest answer is: less than the raw difference, because the people who show up were already the
people most likely to renew.

This page shows how to calculate the difference, how far you can get toward a causal claim, and
what to tell a board that wants a simpler answer than the data supports.

## What does the naive renewal-lift number say?

It says two groups differ. It does not say why.

**Renewal rate (group) = Members in group who renewed within the window ÷ Members in group eligible to renew**

**Naive renewal lift = Renewal rate (attendees) − Renewal rate (non-attendees)**

**Naive attributable renewal revenue = Naive lift × Attendees eligible to renew × Average annual dues**

The third formula is the one that ends up in board decks, and it is the one to be careful with.
It assumes every point of difference between the groups was produced by attendance. Nothing in
the calculation supports that assumption. The presenter imports it.

The naive lift is still worth computing, because it sets an upper bound. Whatever the causal
effect of attending is, it is almost certainly smaller than the raw gap, and knowing the ceiling
is useful when someone proposes spending against it.

## How do you build the cohort so the number means something?

Four decisions, made before you look at the result.

**Fix the population at a date.** Choose a date before registration opens, such as the start of
the membership year, and take every member in good standing then. Anyone who joins later is not
in the cohort. Build the population after the event and you have included people who joined
because of it and excluded people who lapsed before it. Both distortions push the same way.

**Define the renewal window in days.** State it as "renewed within 90 days of the anniversary
date", or whatever your grace period is, and use the same window on both groups every year.

**Define attendance.** A registration is not attendance. Use verified attendance, such as a
badge scan at entry, as set out in the guide to
[event attendance tracking](https://eventiq.io/md/blog/event-attendance-tracking). Track comped and paid attendance
separately. They are different populations.

**Freeze the segment attributes as of the fix date.** Tier, tenure, chapter, job level, and
prior-year attendance must be recorded as they stood before the event. Read them today and you
are reading values the event may have changed, which makes the matching below circular.

Write these four into the report. Half the arguments about renewal lift are arguments about
which population was counted.

## Why is correlation with renewal not causation?

Because the people who attend were selected, by themselves, their employers, and your
marketing, on exactly the characteristics that predict renewal.

Work through who is in the attendee group. They had a travel budget, which usually means a
supportive employer and a stable role. They could take three days away, which means seniority
or control of their calendar. They opened your emails, so they were already engaged. Many had
attended before and liked it. Some are chapter leaders, committee members, or speakers, whose
relationship to the association is structural. Every one of those attributes independently
predicts renewal, and none was caused by the conference.

Now look at the non-attendee group. It contains members who were interested but could not
afford to travel, members who were disengaged, members who had already decided to leave, and
members who were changing jobs. That is not a control group. It is everyone left over.

Three specific ways the naive number misleads:

- **Reverse causation.** A member who has already decided to renew is more willing to spend on
  the conference, so renewal intent drives attendance.
- **Confounding by engagement.** Newsletter reading, directory use, committee service, and
  conference attendance all move together, so crediting the renewal to any one of them
  double-counts.
- **Survivorship.** Members who lapse mid-year often drop out of the eligible population
  entirely, inflating both rates unevenly.

The summary sentence you can put in a report unchanged: attendees renew at a higher rate than
non-attendees, and part of that difference reflects who chooses to attend, not what attending
does.

## How do you get closer to a causal answer?

Three methods, in ascending order of credibility and difficulty. None reaches proof, and you
should not claim otherwise.

**Matched cohorts.** For every attendee, find a non-attendee who looked the same before the
event on the attributes that predict renewal: tenure band, tier, prior-year renewal, prior-year
attendance, chapter involvement, job level. Compare renewal within matched pairs. This removes
the differences you can observe and nothing about the ones you cannot, such as motivation,
career intent, or an employer's plans, which usually matter most. The gap between the naive and
matched lifts tells you how much of the original number was selection.

**Matched renewal lift = Renewal rate (matched attendees) − Renewal rate (matched non-attendees)**

**Selection share of the naive number = (Naive lift − Matched lift) ÷ Naive lift**

**First-time attendee cohorts.** Restrict the analysis to members who had never attended. This
holds prior conference experience constant and isolates a first exposure. It is a cleaner
question and a more useful one, because first-time acquisition is something you can influence.
Selection remains, since something made them come this year, but the strongest confounder is
gone. The guide to [attendee retention](https://eventiq.io/md/blog/attendee-retention) covers the same split from
the event side.

**Natural experiments.** Look for variation in attendance that member motivation did not
drive. Location is the most useful: when the meeting moves cities, travel cost and distance
change for whole segments of the membership for reasons unrelated to engagement. Compare
renewal for members in regions that became cheaper to reach against regions that became more
expensive. Capacity constraints, waitlists, and collisions with other industry events create
similar variation. This is the closest most associations get to causal evidence, so record the
conditions that created it.

A deliberately randomized offer, such as a travel stipend or discounted registration given to
a randomly selected group of eligible members, produces a genuine comparison. It is the only
method here that supports the word "caused", and it has to be planned before the cycle starts.

## What data do you need, and what usually breaks?

One thing, mostly: a member identifier that survives the join between your association
management system and your registration platform. Everything here depends on knowing which
registration belongs to which member record.

It breaks in familiar ways. A member registers with a work email that differs from the one in
the membership system. An assistant registers on their behalf. A member changes employers
between the fix date and the event. A chapter registers a group and the identities arrive as a
spreadsheet. Someone registers as a non-member and is upgraded onsite. Each produces a
registration that cannot be matched, and unmatched registrations do not fail loudly. They
shrink the attendee cohort and push people into the non-attendee group, which biases the lift
upward.

Two fixes are worth the effort. Require member login or a member ID at registration for the
member rate, which is both a data control and a pricing control. And report the match rate in
every cohort analysis: "of 1,800 registrations, 1,642 matched to a member record" tells a
reader how much of the result stands on solid ground.

## Example: a naive lift and a matched lift

Take an association with 12,000 members in good standing on the fix date, average annual dues
of $410, and an annual meeting 1,900 of them attended. All figures are hypothetical and
illustrate the method only.
| Group | Members | Renewed | Renewal rate |
| --- | --- | --- | --- |
| Attended (verified) | 1,900 | 1,748 | 92.0% |
| Did not attend | 10,100 | 7,474 | 74.0% |
| Matched attendees | 1,900 | 1,748 | 92.0% |
| Matched non-attendees | 1,900 | 1,634 | 86.0% |
| First-time attendees | 420 | 378 | 90.0% |
| Matched first-time non-attendees | 420 | 344 | 81.9% |
The pairs are matched on tenure band, tier, prior-year renewal, prior-year attendance, chapter
role, and job level, all as of the fix date.

- Naive renewal lift: 92.0% − 74.0% = 18.0 points. Naive attributable revenue: 0.180 × 1,900 ×
  $410 = $140,220.
- Matched renewal lift: 92.0% − 86.0% = 6.0 points. Matched attributable revenue: 0.060 × 1,900
  × $410 = $46,740.
- Selection share: (18.0 − 6.0) ÷ 18.0 = 67%.
- First-time lift: 90.0% − 81.9% = 8.1 points.

Three readings. Two thirds of the headline number was selection: matched non-attendees renewed
at 86%, not 74%, because they looked like attendees on everything measurable except attending.
The defensible revenue figure is $46,740 and not $140,220, and a skeptical CFO will find that
difference if you present the larger number. And the first-time cut shows a larger lift than
the matched average. If that holds across two cycles, it argues for spending on first-time
acquisition over general attendance promotion.

Note what the matched figure still does not establish. It does not prove the meeting caused 6.0
points. It says that after removing every difference you could observe, a 6.0-point gap
remained, and that gap could still be driven by something unmeasured.

## What to do this quarter

- Write down the four cohort decisions (fix date, renewal window, attendance definition, frozen
  attributes) and publish them with the result.
- Compute the naive lift and label it as an upper bound.
- Build matched cohorts on the attributes you already hold, and report the selection share
  beside the matched lift.
- Add a first-time attendee cut. If the population is too small to match, say so and do not
  publish a precise-looking number.
- Report your member-ID match rate in the same table as the result.
- Record location, capacity, and scheduling variation so you have a natural experiment next
  year.

## Common questions

### Is the matched lift the true causal effect?

No. It is the naive lift with observable selection removed. Unobserved differences, such as
motivation, career stage, and employer support, remain and generally push the same way, so the
matched lift is still more likely to overstate than understate the effect.

### Should we use an engagement score instead?

An engagement score is useful for targeting and poor for causal claims, because attendance is
usually one of its inputs. If attendance feeds the score and the score predicts renewal, you
have built a circle. Keep attendance out of any score used as a control variable.

### How do we handle members who attend on a comped badge?

Track them as a separate group. Comped attendees were selected by someone in your organization,
which is a different selection process from self-funded attendance, and mixing them in makes
the matched comparison harder to defend.

### What should we tell the board?

Three numbers and one sentence: the naive lift as a ceiling, the matched lift as the defensible
figure, and the selection share explaining the gap. The sentence is that attendance is strongly
associated with renewal, that part of that association reflects who attends, and that the
conference reaches the members most likely to stay either way, which may still be the right
investment. The [board report template](https://eventiq.io/md/templates/board-report) has a place for all three.

## Where EventIQ fits

EventIQ replaces nothing. It connects on top of the platforms you already run: event platforms
(Cvent, Zoom, Swapcard), CRM (Salesforce, HubSpot, GoHighLevel), and marketing (Google Ads, Meta
Ads, LinkedIn Ads, Mailchimp, Google Analytics). Platforms with an API outside that list are
connected on request.

Cohort work fails on plumbing more often than on method, and the membership half of that
plumbing is outside the product today: there is no connection to an association management
system, so membership and renewal records still come out of your membership system by hand.
For [associations](https://eventiq.io/md/for/associations), that is the limit to know before a demo.

What EventIQ holds is the event half. Registrations and attendance arrive as separate records,
and attendees are matched to CRM contacts by exact email, with the result kept on every record:
matched, unmatched, or no email. It does not invent a link where the identifier is missing. An
unmatched registration stays unmatched.

So the cohort table on this page is a join your team still performs, using EventIQ for the
attendance half and your membership system for the renewal half.

[Book a demo](https://eventiq.io/#early-access) to see registrations, attendance, and the match result on each
record on a sample event, in a 20-minute demo.

---

HTML version: https://eventiq.io/blog/member-retention-event-attendance