HSOPSC items

Item Difficulty (Location Parameter) Each HSOPSC item has five possible response options and thus four location parameters. These are estimates indicating at what point along the patient safety construct that the respondent will shift from one option (say a 1 ) to the next option (say a 2 ) on the Likert scale. This location, then, determines the point where the item s severity in assessing safety culture equals the individual respondent s level of safety culture. 

The MULTILOG 7 software program (Thissen 1991) was used to fit a graded IRT response model (Samejima 1969) estimating item discrimination and location parameters (Embretson and Reise 2000) for each HSOPSC item. SPSS software (PASW 18.0) was used to calculate correlations between mean scale scores and corresponding IRT scale scores. 

Table 8.2 displays HSOPSC discrimination (a), as well as the four location parameters b, b, b, and b ) for each item. Note that HSOPSC items labelled with an r are reverse-scored. Items with poor discrimination (i.e. a < 1.35) are highlighted. These items (B3r, B4r, A9, All, A5r, A7r, F6r and A15) match those that were dropped, based on the results of an exploratory factor analysis conducted on data for the present study (Waterson et al. 2010), which highlights that the slope parameter is analogous to a factor loading (item-factor correlation) in factor analysis. In addition, item FI Ir Shift changes are problematic for patients in this hospital also had low discriminating power. 

Comparing the ICCs for both items shows that the response options of the HSOPSC item in Figure 8.3a are sensitive (a > 1.35) and will discriminate between individuals with different levels of safety culture perceptions, except for those who choose Strongly Disagree versus Disagree because the estimated difficulty parameter is located below -3 standard deviations. In contrast, the response options of the HSOPSC item in Figure 8.3b are not sensitive (a < 1.35). 

The third location parameter, b, discriminates between those people who Neither Agree/Disagree and those who Agree . Estimates for b ranged fiom -2.6 to +0.3 for HSOPSC items with slope parameters above 1.35. This implies that even people who are about 3 standard deviations below the mean are likely to at least Agree on a 5-point Likert scale. For the Number of Events Reported b parameter was +0.7... 

This chapter provides an overview of the psychometric properties of the HSOPSC using both classical test theory (CTT) and the modem approach, often referred to as Item Response Theory (IRT). To enhance the understanding and importance of IRT, the basic principles will first be introduced. In partictrlar, three fundamental outcomes of the IRT approach will be highlighted (1) item characteristic curves, (2) measurement information, and (3) invariance. Moreover, this chapter will present data that have been previously analysed and... 

In order to tailor the HSOPSC to a UK healthcare setting, a nirmber of changes were made to the original version of the instrument. Specifically, in 11 items, the expressions area and unit were changed to ward and department for five items error and mistake were replaced by adverse outcome over and over was replaced by repeatedly for one item two items were excluded based on feedback from hospital management arrd change of item meaning (see Waterson et al. 2010, for more details). 

Table 8.2 Item parameters for each HSOPSC dimension and ontcome measnre (correlation between IRT-scoring and original mean-scale-scoring)...

In conclusion, this chapter demonstrates that the IRT approach can provide additional insights to psychometric properties of the HSOPSC. Both, the classical and modem approaches, are needed to form a complete picture of the properties of a set of items. Understanding the basic principles of IRT will hopefully foster its use more widely within the field of patient safety culture assessment. This will ultimately enhance our ability to measure this important constmct accurately. 

---