Table 1. Measurement properties and criteria assessed in the Rasch analysis.

Measurement properties	Purpose	Statistical test	Measurement criteria
Unidimensionality and local dependency	To assess whether items in each SAQ-INA domain measured the same underlying construct	• PCA of residual • Equating t-test • Binomial dimensionality test	Identify two subsets of items from the first factor extracted by PCA [26]. Compare person estimates from the two different subsets using independent t-tests, where p<0.05 indicates the domain is unidimensional. If p>0.05, the value of 5% obtained from a binomial test of proportions should be included in the 95% CI [26].
	To assess whether the response to an item is dependent on the response to another item (i.e. local dependency which is an element of unidimensionality).	• Person-item residual correlation	A person-item residual correlation value of >0.2 indicated the presence of local dependency [26].
Response format (thresholds)	To assess whether participants could discriminate between the different response options of the five-point Likert scale.	• Threshold map • Category probability curves	Visually inspect the pattern of response options for each SAQ-INA item. Thresholds are considered to be ordered when each response option is the most likely response at some point along the location continuum [26].
Targeting	To assess whether items in each SAQ-INA domain had floor or ceiling effects.	• Mean location score • Person-item distribution threshold distribution map	A mean logit score of zero indicates a well-targeted scale i.e. no floor or ceiling effects [26]/
Internal consistency reliability	To assess whether items in each SAQ-INA domain can differentiate varying levels of safety climate.	• Person separation index	A value of >0.7 suggests that the SAQ-INA items has good internal consistency reliability [16, 26].
Item bias	To assess whether items in each SAQ-INA domain were biased towards specific groups (e.g. public versus private hospitals).	• Differential item functioning (DIF)	Significant main (uniform DIF) and interaction (non-uniform DIF) effects (p<0.05) indicates that an item may be biased for different groups e.g. public versus private hospitals [26].

SAQ-INA: Safety Attitudes Questionnaire–Indonesian version; PCA: Principal Component Analysis; CI: Confidence interval