Formative assessment is essential for improving instructional quality and supporting student learning, yet valid and contextually appropriate instruments for measuring teachers’ formative assessment practices remain limited in West Africa. This study examined the dimensional structure and psychometric properties of the Teacher Formative Assessment Practices Scale (TFAPS) among in‑service teachers in West Africa and identified the most appropriate multidimensional item response theory (MIRT) model for ordered‑category data. Using a cross‑sectional design, data were collected from 182 teachers in Ghana and Nigeria via an online survey. The TFAPS comprises 10 Likert‑type items representing two theoretically grounded dimensions: teacher‑directed and student‑directed formative assessment practices.

To evaluate dimensionality, ordinal confirmatory factor analyses were conducted using maximum likelihood and weighted least squares mean and variance adjusted (WLSMV) estimators. Both two‑factor and bifactor models demonstrated acceptable global fit; however, the bifactor model estimated with WLSMV showed superior fit, supporting a multidimensional structure with a substantive general factor. Based on this structure, multidimensional Partial Credit Models (PCM) and Generalized Partial Credit Models (GPCM) were estimated and compared using information criteria, limited‑information fit indices, item‑level statistics, and reliability estimates.

Results consistently favored the GPCM, which demonstrated better global and item‑level fit, higher incremental fit indices, and slightly improved person reliability across both dimensions. Additionally, the GPCM yielded lower latent factor correlations, indicating clearer separation and enhanced interpretability of the constructs. Wright Map analyses revealed disordered thresholds and limited alignment between item difficulty and teacher ability, suggesting that the six‑category response format may be overly granular. Overall, findings support the TFAPS as a psychometrically sound instrument in the West African context, with evidence favoring the GPCM and indicating the need for response‑category refinement to improve measurement precision and practical utility.