Description
Compare groups' mean profiles across several measures on the same scale by testing in order whether profiles are parallel, at the same level, and flat.
When variables are commensurate, sharing units and similar variances (a battery of rating attributes on the same scale, or repeated measurements), each group's mean profile is the line joining its variable means. Profile analysis asks three ordered questions using contrasts of successive differences. First, parallelism: are segment-by-segment changes equal across groups, meaning no group-by-variable interaction? This is a Hotelling T-squared test on the contrasts. Only if profiles are parallel do the next two make sense: equal levels, whether one group scores uniformly higher (a t-test on summed scores), and flatness, whether the combined profile is level across variables. If profiles are not parallel, the interaction is interpreted from the plot. One-sample profile analysis tests flatness of a single mean vector, and several groups are handled by MANOVA on contrasts.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Siloed
- Stage
- Discovery, Evaluation
- Data
- Quantitative
- Effort
- Hours
Steps
- Confirm the variables are commensurate: same scale and broadly similar variances.
- Form successive-difference contrasts across the variables.
- Test parallelism with Hotelling T-squared (or MANOVA for more than two groups) on the contrasts.
- If profiles are parallel, test equal levels with a t-test on each person's total.
- If profiles are parallel, test flatness of the combined profile.
- If profiles are not parallel, plot them and interpret where the group-by-variable interaction lies.
Sources
- Methods of Multivariate Analysis (Rencher, 2002)
Also known as
- Two-sample profile analysis
- Test of parallel profiles
- Parallelism test
Categories
- Analytics & Behaviour
- Statistics & Modelling