Studio equipment attribution methods: four ways to link kit purchases to outcomes, and how strong each conclusion is

What to take away

Counterfactual impact design is the only method here that can carry a causal claim, and only with a credible comparison group.
Tagged-source allocation gives a descriptive count of recorded routes.
Respondent source statements recover influences that logs miss, but they remain self-report.
Contribution assessment tests a theory and returns a bounded judgement, not a causal percentage.

The four methods on one page

Buying a camera, microphone or lighting kit is a cost, and attribution asks what that kit changed.

All four methods compare the same unit: an eligible downstream event per eligible tutorial released, inside one observation window. Use the same event and the same window for each, or the results cannot sit side by side.

Method	Outcome unit	Data need	Main bias	Privacy burden	Causal ceiling
Tagged-source allocation	Events allocated to a source	Stable identifier, fixed window, duplicate rule	Coverage gaps, cross-device loss, bots	Consent state, storage and access assessment	Descriptive count
Respondent source statement	Eligible events, self-declared	Fixed question, options, set timing	Recall, question order, social desirability	UK GDPR lawful basis, anonymity choices	Declared routes only
Contribution assessment	Bounded judgement	Documented chain, rival theories, contrary evidence	Confirmation pressure on the analyst	Mostly documents, low personal data	Evidenced reasoning, no percentage
Counterfactual impact design	Estimated causal change	Pre-specified comparison, assignment, analysis	Selection, interference, attrition	UK GDPR lawful basis for assignment data, retention limits	Causal estimate with uncertainty

Example: one softbox and a lighting tutorial

A two-person studio buys one softbox and publishes a tutorial based on its lighting kit choice.

Each method answers a different question about the same launch. The same logic applies to a new camera, microphone or lens: the purchase is the intervention, and the downstream event is what you try to link to it.

Tagged-source allocation counts tutorial visits that arrive through the launch link. It cannot say the softbox caused them.
A respondent statement records what viewers remember influencing them, and may recover a newsletter mention the logs lost.
Contribution assessment checks whether the expected steps happened and names the rival explanations still standing.
Counterfactual design compares studios that did and did not receive the softbox. Only this route estimates the change.

Tagged-source allocation: data, bias and privacy under PECR

You need stable identifiers, a fixed window, a duplicate rule and a written route from release to event. The output is a count allocated to a recorded source under that rule. It is not the cause of the event. Our measurement and reporting framework shows how to define those fields.

Coverage is the main bias. Missing consent states, cross-device activity and automated traffic shift the count without touching the equipment. See measurement mistakes for the errors that skew an upgrade verdict.

Tracking tools need a privacy check, and that check covers cookies, tracking pixels and link decoration, among other technologies (see the storage and access guidance).

Respondent source statement: routes the logs miss

Ask eligible viewers how they first met the tutorial, with a fixed question and fixed options. Keep "do not know" and non-response visible in the data.

Recall, question order and social desirability shift the answers, and a small pilot is the only way to estimate the size of each shift. Our audience research guide shows how to write the question set and record consent.

Contribution assessment: testing the chain

Write the chain from equipment change to production behaviour to the defined event. Then test it against rival chains.

Write the proposed chain and the competing chains.
Gather evidence for and against each step.
Record what would make the chain fail.
Publish the judgement with its assumptions and gaps.

Never convert this output into a percentage of causal credit.

Counterfactual impact design: the causal ceiling

Pre-specify the intervention, the eligible units and the outcome. Then set the comparison, assignment, period and analysis.

The Magenta Book on evaluation frames counterfactual evaluation as asking what would have happened without the intervention, and how far observed changes can be attributed to it.

Report the estimate with its uncertainty, any departures from plan and limits on which studios or outputs it covers. A small estimate is not directional proof. A check on benchmark research matters when comparison studios differ in size or output mix.

Matching the methods to kit changes

Pick the outcome before the purchase, not after. The table pairs common kit changes with an outcome you could define and the method to start with.

Kit change	Outcome to define	Method to start with
Camera or lens upgrade	Eligible tutorial views in the window	Tagged-source allocation
Microphone upgrade	Viewers who name audio as a reason to stay	Respondent source statement
Lighting kit	Completion of the lighting tutorial	Contribution assessment
Studio-wide kit refresh	Change in the agreed outcome against a comparison group	Counterfactual impact design

Choose by the decision

Use allocation for a count, respondent evidence for declared routes, contribution work to test a theory and counterfactual evaluation for a proportionate causal question. Record the choice before you view results.

Define the same event and window across methods.
Record the method choice before results.
Keep "do not know" and multiple influences in survey data.
Do not convert contribution work into causal credit.
Report causal estimates with uncertainty and plan departures.

Never average the outputs of different methods. If tagged and self-reported routes disagree, keep both and investigate coverage rather than choosing the larger figure. None of the four methods measures device safety, audio quality or audience welfare, so use a selection checklist for those.