The consent line, and the bias it leaves behind

A consent banner is not only a legal step; it is a filter on the data. The readers who decline are removed from the denominator, and if they behave differently the reported rate describes the readers who stayed rather than the readers who came. On the samples the difference is 5.10% against 2.30%.

Desk spec

accept analytics: 68.4%
unmeasured: 31.6%
measured rate: 5.10%
population rate: 4.22%

the eventOne interaction, written down. A click carries a name, a time, an account, a session and what was on the screen, and it is kept whether or not the reader chose to be measured.

the funnelThe order the steps happen in. A landing visit becomes an account, an account becomes a deposit page, and 100,000 visits end in 2,074 first bets - a 2.1% path the whole loop is aimed at.

the armsTwo versions of one screen shown at the same time, and a rate for each. 4.30% against 5.16% is a 0.86-point lift, and the interval decides whether it is a result or a coincidence.

Direct answer

A consent line decides who is measured. Readers who decline non-essential analytics are excluded from the event log, so every rate computed from it is a rate on the consenting group. On the samples the consenting 68.4% behave at 5.10% and the unmeasured 31.6% at 2.30%, so the true population rate is 4.22% and the dashboard overstates it by 20.9%.

Who is removed, and whether it matters

Excluding a random slice of visitors would change the rate by nothing. Excluding a slice that behaves differently changes it by everything, and the readers most likely to decline are usually the more cautious ones - which on a gambling site means the ones least likely to deposit.

Sample H - 100,000 visits, split by what the banner produced
Group	Visits	Share	Deposit rate
accepted analytics	68,400	68.4%	5.10%
declined non-essential	21,600	21.6%	2.30%
made no choice	10,000	10.0%	not measured
unmeasured in total	31,600	31.6%	2.30% assumed, from the decliners

sample H - the measured rate against the population rate measured rate, on the 68.4% who accept = 5.10% assumed rate on the 31.6% who do not = 2.30% population rate = 0.684 x 5.10 + 0.316 x 2.30 = 3.4884 + 0.7268 = 4.2152 -> 4.22% overstatement = 5.10 / 4.2152 = 1.2099 -> 20.9% so a dashboard reading 5.10% is describing a population whose real rate is 4.22%, and no page in the product says so. the two arms of sample A were both measured on the consenting group, which is why the lift survives the bias even though the level does not.

What is measured with and without a choice

Consent is not a single switch over all recording. Some of it is treated as necessary to deliver the page and some as optional, and the line between the two is drawn by the operator and by the law where the reader is.

Sample H - the categories, and what each is used for
Category	Needs a choice?	What it produces
a session identifier to keep a login working	no, necessary	the account stays signed in
a security record of a device	no, necessary	the signal behind a takeover check
a page view counted in aggregate	usually yes	the top of the funnel
a clickstream tied to an account	yes	the funnel's middle steps
a session replay	yes, and rarely named	the recording described on this site
cross-site advertising data	yes	nothing the desk's loop needs
the loop's own inputs	yes	the whole experiment depends on a choice the reader may refuse

sample H - what the refusal costs the measurement visitors who refuse or never choose = 31.6% deposit rate among them, assumed = 2.30% deposits the dashboard never counts = 31,600 x 0.023 = 727 deposits it does count = 68,400 x 0.051 = 3,488 so 727 of 4,215 deposits, or 17.2%, are invisible an experiment run only on the visible 3,488 can move the measured rate by a fifth of a point without moving the real one at all, and the holdout of sample E is the only part of the loop that would notice.

The rates are invented and the legal column is deliberately general, because the rule differs by country. What is not general is the arithmetic: a group that refuses is a group that is not counted, and a rate computed without it is a rate on the remainder. A reader who declines should assume they are simply absent from the numbers rather than represented in them.

What a consent choice does and does not do

Declining removes you from the counts; it does not remove the events already recorded.
Declining usually applies going forward, not retroactively.
A necessary category is defined by the operator and can be broader than a reader expects.
A replay may be bundled into an analytics category rather than named separately.
Changing the choice is usually possible, and worth doing if the first one was made under pressure.

The consent line, and the bias it leaves behind

Who is removed, and whether it matters

What is measured with and without a choice

Read next