A Practical Approach to Develop a Parsimonious Survey Questionnaire—Taiwanese Patient Safety Culture Survey as an Example

This study aims to suggest a method to choose what items can be removed from a survey instrument, thereby lessening survey participants’ burden. We used the Taiwanese Patient Safety Culture (TPSC) survey as an example—in particular, the emotional exhaustion (EE) and work–life balance (WLB) domains. A traditional factor analysis approach, whereby a Likert scale is treated as a linear variable, was applied, and an item response theory (IRT) graded response model (GRM), where a response is treated as an ordinal scale, was used. From both methods, five out of nine EE items and three of seven WLB items were filtered out for removal; these items provided significantly smaller information for domain score estimation than other items. To check whether the downsized versions showed sufficient validity, we tested the factor structure of the remaining items and compared fit indices between the models with all original items and the new models with the remaining items. The new models in both EE and WLB domains showed better model fit than the models with all original items. In this study, we showed how to select more important items while ensuring that the selected items as a whole can provide a better fit compared to the original instrument. This interesting phenomenon may be due to the translation step and different cultural soils where surveys are administered. We strongly recommend checking the possibility that items can be removed when adopting and validating a survey instrument—not only to shorten the overall length, but also to increase validity.