Description
Log how much each participant in an intervention actually engages, relate that dose to outcomes within the treated group, and check baseline measures do not predict later engagement.
Intervention trials often show weak group effects because many participants barely use the intervention. Dose analysis records actual practice or engagement for each person in the treated arm and tests whether more use goes with better outcomes, as a continuous relationship or a median split. In a compassion training trial there was no group effect on stress hormones, but high-practice participants recovered to baseline cortisol faster and produced less of an inflammatory marker. The key safeguard is a reverse-causation check: if baseline outcome measures do not predict later practice, self-selection becomes a less likely explanation. The association remains correlational because dose is self-selected, so it is reported alongside, not instead of, the between-group comparison. In digital products, it maps directly onto engagement with coaching, learning or health apps.
Attributes
- Discipline
- Experimentation & Causal Inference
- Loss type
- Latent
- Stage
- Evaluation, Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Log practice or engagement per participant throughout the intervention (sessions, minutes, completed exercises).
- Run the planned between-group comparison as the primary analysis.
- Within the treated arm, test the dose-outcome association, continuously or with a median split.
- Check that baseline outcome measures do not predict later engagement, to argue against self-selection.
- Report dose-response estimates alongside the between-group result, labelling them as correlational.
Sources
- Compassion: Bridging Practice and Science (Singer & Bolz, 2013)
Also known as
- Within-group practice-dose analysis with a reverse-causation check
- Practice-time dose-response analysis
- Engagement-dose analysis
- Median split on home practice
Categories
- Analytics & Behaviour
- Experimentation & Causal Inference