Analysis of Covariance
A Multiphase Optimization Strategy (MOST) factorial analysis. Adapted from code from MS and based on Spring et al. (2020).
Overview
Design
The study uses a 24 full factorial design with four binary engagement strategies:
| Factor | Code | Off/No | On/Yes |
|---|---|---|---|
| Gamification | Gamif |
No gamified content | Gamified content enabled |
| Health Coach | HealthCo |
No coach access | Access to health coach |
| Parent Resources | ParRes |
No parent resources | Parent resources provided |
| Text Messages | TextMes |
No text messages | Text messages sent |
All factors are effect coded (Off/No = −1, On/Yes = +1). This coding means each main-effect coefficient represents half the difference in EI between the On/Yes and Off/No groups, averaged across all combinations of the other factors.
Analysis sequence
Following Spring et al. (2020):
- Full ANCOVA — fit all main effects and interactions simultaneously
- Screening — identify components with main effects at α = .10 (liberal, to reduce Type II error)
- Interaction testing — examine higher-order interactions among screened-in components
- Parsimonious model — refit retaining only screened-in effects
- Inference — report p-values with α = .01 (per protocol)