"The level of collapse we are allowed: Comparison of different response scales in Safety Attitudes Questionnaire"

Abstract

Various survey instruments have been used to measure patient safety culture. Many of these instruments use a (usually 5-point) Likert scale. This study used the Safety Attitudes Questionnaire-Korean version (SAQ-K), consisting of 34 items in six domains, to examine whether other scales, such as dichotomized and trichotomized scales, present equivalent estimates to the currently used 5-point Likert scale (1=disagree strongly, 2=disagree slightly, 3=neutral, 4=agree slightly, 5=agree strongly). For each item, we generated a 3-point scale by collapsing responsesfor 1 and 2 into one category and 4 and 5 into another category, yielding a scale of 1=disagree, 2=neutral, and 3=agree. A dichotomized scale was generated by collapsing responses for 1 through 3 from the original scale to 0=disagree and 4 and 5 to 1=agree. Correlations among the results from the five measurement scales for each respondent, as the unit of analysis, for each of the six domains were estimated: currently used simple mean of item scoreson a 5-point scale, empirical Bayes (EB) estimate from a 5-point graded response model (GRM), EB estimate from a 3-point GRM, EB estimate of a 2-parameter (2PL) item response theory (IRT) model, and EB mean of a 1-parameter (1PL) IRT model. All correlation coefficients were statistically significant (p<.01) and mostly exceeded 0.9 between the currently used simple mean and 3-point GRM estimates, although for dichotomized scales most coefficients were between 0.8 and 0.9. When we aggregated the responses to the clinical area level, the correlation became much higher, exceeding 0.9, except for those involving dichotomous scales in the stress recognition domain. This study found that dichotomous or trichotomous scales performed well compared to the current 5-point scale, suggesting such collapsing could replace the original scale at least in the analysis phase of collected data. Further study is needed to examine whether such simpler scales can be used in the survey-administering phase with sufficient validity.