Pragmatic Solomon Four-Group Design — Real-World Pretest Sensitization Control

The main barrier is sample size. Splitting participants into four groups rather than two roughly doubles the required N for equivalent power in the primary treatment comparison, making the design expensive. Researchers often judge the pretest sensitization risk as low enough to tolerate, especially when baseline measures are physiological rather than attitudinal. The design is most justified when the intervention targets the same attitudes or behaviors that the pretest directly measures.

Both use the same four-group structure to control for pretest sensitization. The pragmatic version applies broad eligibility criteria, delivers the intervention through routine practice channels without a scripted protocol, uses patient-relevant or system-relevant outcomes, and operates with a light compliance and fidelity monitoring framework. The explanatory version applies strict eligibility, highly controlled delivery protocols, and is primarily interested in efficacy under ideal conditions. The pragmatic version sacrifices some internal control to gain external validity and direct generalizability to real practice.

The standard approach is a 2x2 factorial ANOVA (or ANCOVA if covariates are included) on the posttest scores of all four groups: Factor A = intervention status (yes/no), Factor B = pretested (yes/no), and their interaction A x B. A significant interaction indicates that pretest sensitization occurred. If the interaction is non-significant, the simpler main effect of Factor A is the primary estimate of the treatment effect. Mixed-model or multilevel approaches are preferred when cluster randomization is used, as in most pragmatic settings.

Yes. A delayed posttest (e.g., 3 or 6 months after intervention) can be added to all four groups to assess sustainability of effects. However, by the follow-up time point, Groups 3 and 4 still have no baseline, so individual-level change analysis remains limited to Groups 1 and 2. The factorial ANOVA framework extends naturally to include the additional time point as a repeated measure.

Yes, and cluster randomization is often the natural choice in pragmatic settings where providers or sites — rather than individuals — deliver the intervention. Clusters (schools, clinics, wards) are randomized to one of the four cells. The analysis must account for within-cluster correlation using multilevel models or GEE; standard ANOVA will produce anti-conservative standard errors and inflated Type I error rates.