Skip to main content

Table 1 Statistical tests and criteria for assessment to examine specific measurement properties of the SAQ

From: Measuring safety climate in acute hospitals: Rasch analysis of the safety attitudes questionnaire

Measurement property Purpose Statistical test Criteria for assessment
Unidimensionality To assess whether items in each SAQ domain measure a single construct (or concept) • Residual fit statistics
• Item-fit residuals
• Person-fit residuals
• A fit residual SD value >1.5 would suggest a problem [18].
• Item residuals that range between −2.5 and 2.5 indicate adequate fit to the model [18].a
Response formats (thresholds) To assess whether participants had difficulty discriminating between the response options on the SAQ • Threshold map
• Category probability curves
• Pattern of thresholds examined.
• Ordering of thresholds where each response category systematically take turns to be the most likely response [18].
Targeting To assess whether the 6 SAQ domains are appropriately targeted for the clinical population (floor and ceiling effects) • Mean location score
• Person-item threshold distribution map
• The mean sample location should approximate the mean item location (i.e. zero) for a well-targeted measure [18].
Internal consistency reliability To assess the extent to which items distinguish between levels of safety climate • Person separation index (PSI) • A PSI of a > 0.7 indicates the items of the scale is able to separate the participants in the sample [24].b
Item bias To assess whether different groups within the sample (e.g. medical or surgical ward and nursing qualification), despite equal levels of the underlying characteristic being measured, responds in a different manner to an individual item • Differential item functioning (DIF) • Uniform DIF is indicated by a significant main effect for the person factor (e.g. ward type) [18].
• Non-uniform DIF is indicated by a significant interaction effect [18].
  1. aItems with large negative residual values indicate a high level of predictability in responses and signal possible item redundancy. Items with large positive residual values suggest an item does not contribute to the measurement of a unidimensional construct
  2. bA PSI is the same as Cronbach’s alpha with the logit value replacing the raw score in the same formulae