Steps | Psychometric property | Aim | Criterion |
---|---|---|---|
1 | Rating scale function | Assess the scale’s functionality, i.e. do the category measures on each item advance monotonically | Goodness-of-fit: < 2.0 outfit MNSQ, minimum 10 participants per value per item |
2 | Internal scale validity | Examine how well the item responses match the expected responses in the Rasch model | Item goodness-of-fit: ≤ 1.2 MNSQ, worst fitting item removed one at a time and model subsequently re-run |
3 | Dimensionality | Assess if the scale measures a single construct | > 50% total variance explained by 1st component (Rasch model), additional components ≤ 5% (or eigenvalue ≤ 2.0) after removal of first component. No more than 1 out of 20 (or 5%) of the residual correlations > 0.30 |
4 | Reliability | Person-separation validity: Assess if the scale can discriminate participants’ responses into groups based on performance; Internal consistency: Assess if the item responses are consistent | Person-separation index: ≥ 2.0 Internal consistency: Cronbach’s alpha ≥ 0.80 |
5 | Differential item functioning (DIF) | Examine how the scale functions among various groups (number of LTCs, age, gender, cohort, hospital admissions) | DIF contrast < 0.43 logits: p > 0.01 |