Read a change without changing the denominator

Fictional example: two comparable periods, each with 1,000 visits and the same definition of a registration. These numbers illustrate a calculation, not measurements of a real network.

Registrations per 1,000 visits
PeriodRegistrationsRate
A10010%
B12012%

The rate increases by 2 percentage points. Registrations increase by 20%: (120 − 100) ÷ 100. Calling this “+20 points” would be incorrect.

If the duration, tracking or paid traffic share changes, explain those differences before drawing conclusions. A session-based rate cannot be compared directly with a person-based rate.

Why not automatically call this an engagement rate?

Services define their own metrics. In GA4, engagement rate is engaged sessions divided by sessions; it is not the registration rate in this example.

Official definition: Google Analytics ·

Paper charts and a calculator for examining measurements — AI-generated photograph
AI-generated photograph · Fictional scene. Credits

Connect measurement to a decision

For filling a workshop, enquiries and registrations are more directly useful than likes. For a tutorial, identify unanswered questions and misunderstood steps. Define an action you could change based on results; an indicator without an associated decision may simply encourage habitual checking.

Read the service definition

Services define and count indicators differently. YouTube describes impressions click-through rate using registered impressions, not every possible appearance of a video. Read scope, period and exclusions before comparing. The same word in two dashboards does not guarantee an identical method.

Keep comparisons honest

Compare comparable periods and formats. Note frequency changes, advertising, external events and audience size. An increase after a change does not prove the change caused it. Avoid conclusions from one post. Keep counts with percentages so variation based on few responses does not appear universal.

Produce a short useful note

Summarise the question, observations, limits and next action. Do not collect individual profiles to supplement an aggregate measure without an established need. Missing data remains unknown: do not replace it with an estimate presented as real. State when measurement will be reviewed.

Choose the right check for the situation

These editorial pointers help prepare and review your work. Choose the situation matching your content; check current features and conditions in the cited sources before applying it to a service.

Situations, useful actions and mistakes to avoid
SituationUseful actionMistake to avoid
Comparing two periodsUse comparable durations and record format or frequency changes.Comparing a very active week with a full month without context.
Increased visibilityCheck what the metric counts and whether it covers people or displays.Automatically turning impressions into distinct people.
Registration objectiveFollow the journey through confirmed registrations and distinguish clicks from conversions.Declaring success because a post receives many reactions.
Small sampleShow counts, period and limitations; wait before generalising.Presenting a change of a few responses as a lasting trend.

Hypothetical example: apply the method

A team compares two workshop announcements. It records questions and registrations, plus differences in date and format. It concludes that some information was clearer without claiming to have isolated a causal effect.

Template to adapt

Copy this outline into your working document. It collects no information on this site; complete it for your situation.

Question:
Indicator and definition:
Period:
Counts:
Observations:
Limits:
Next action:

Your pre-use check

Checks for this method

0 / 4 checks completed

Frequently asked questions

Can views on two networks be compared?

Check definitions and conditions. Direct comparison may mislead when counting rules differ.

Which number proves success?

No isolated indicator answers every objective. Define the useful outcome and measurement limits.

Continue with related content

Measure results and explain data

Compare indicators answering the same question

Fictional scenario and editorial method; no platform outcome is measured here.

An association gets more views but does not know whether its announcement helps fill a workshop.

Prepare the trial

Define the useful action, period and counting method before reading results. Separate views, people, clicks and confirmed registrations; record missing data.

Observe a useful outcome

The table explains each value and connects observation to a decision. Other contributing channels are identified.

When to reconsider your approach

If definitions or periods differ, do not create a ranking. Unknown data remains unknown rather than becoming zero to complete a table.

Sources and scope

Primary references consulted for factual points. Methods and templates are editorial suggestions; service conditions should still be checked in current documentation.

A question before putting this into practice

Can audiences from multiple networks be added together?

Avoid presenting that sum as unique people: definitions differ and audiences may overlap. Retain each metric with its platform, period and unit. For decisions, compare useful actions and effort using the same explicitly described method.

One number, two interpretations

Fictional example over the same period: a widely displayed post can lead to fewer useful actions. These values illustrate a calculation and do not describe any platform.

One number, two interpretations
PostImpressionsClicksRegistrationsClicks / impressionsRegistrations / clicks
Post A10 000200202 %10 %
Post B5 000120302.4 %25 %
Registrations compared on one common scale
Post A20
Post B30

A receives more clicks, but B receives more registrations. The denominator matters: 30 registrations from 120 clicks is 25%, not 0.6%. None of these calculations establishes causation.

Explore a conversion rate

Rate : 25 %

The calculation does not deduplicate people. Define the action, period and counting method before comparing results.