Within-Country Comparability of the PISA School Belonging Scale, 2018-2022: Invariance Diagnostics and Latent Mean Change for 69 Education Systems

PISA reports how students’ sense of belonging at school changed between 2018 and 2022 in each education system. This data descriptor presents an atlas that tests, system by system, whether the six items behind that index measure the same two constructs in both cycles, for the 69 systems with at least 500 respondents in both. The items fit two factors separated by wording (CFI 0.997, RMSEA 0.041; one factor: 0.906 and 0.214). With WLSMV and the conventional cut-offs, scalar invariance holds in 65 systems and fails in France, Macao, Montenegro and Viet Nam. Freeing one intercept restores comparability in the first three; Viet Nam fails already at the threshold step, needs two freed items, and its estimates are not recommended. With a stricter cut-off, 20 systems fail, including all 7 atlas systems to which the OECD gave version-specific item parameters. Changes go in both directions and are mostly small (median absolute change 0.078 and 0.069 standard deviations on the two factors), with replicate-based standard errors, and in the invariant systems they are closely correlated with the OECD index (r = 0.84 and 0.93). The 33 files, the scripts and a data dictionary are deposited at Zenodo.