Rasch statistic | Fit criteria |
---|---|
Overall fit of items, mean and SD: To what degree do observed item responses correlate with expected responses from the Rasch model? | Perfect fit = mean of 0 and SD of 1 Acceptable fit = SD < 1.5 |
Overall fit of persons, mean and SD: To what degree do observed person responses correlate with expected responses from the Rasch model? | Perfect fit = mean of 0 and SD of 1 Acceptable fit = SD < 1.5 |
Item-trait interaction, chi-square probability value: What is the probability that the overall responses fit the model? | Nonsignificant Bonferroni-adjusted probability value [38, 48] * |
Person separation index: To what degree are item responses consistent across respondents? | Values ≥0.8 [49] |
Thresholds: Do the response categories and thresholds work as intended, i.e. advancing monotonically, or are there any disordered thresholds? | Ordered thresholds [38] |
Individual item fit: To what degree do the observed individual item responses correlate with expected responses from the Rasch model? | Fit residual = +/− 2.5 [50] Chi-square probability values nonsignificant with Bonferroni-adjusted probability value [38, 48] * Visual check of item characteristic curves in which the observed values should fit as closely as possible to the theoretical curve [38] |
Individual person fit: To what degree do the observed individual responses correlate with expected responses from the Rasch model? | Fit residual = +/− 2.5 [50] |
Targeting: To what degree are items targeted to the persons in the sample? | Mean location score for the persons should be close to the mean value of zero set for the items [51] |
Differential item functioning: Does any item deviate from the a-priori requirement of invariance across groups for gender, age and care? | Nonsignificant Bonferroni-adjusted probability value [38, 48] * |
Local dependency: Does any item show dependency on a response to another item? | No positive residual correlations > 0.2 above the average residual correlations across all items [52] |
Dimensionality: To what degree does the questionnaire measure one single dimension? | The proportion of t tests reaching significance should not exceed 5% in the independent t test protocol [53, 54] |