3.1. Respondent Characteristics
Table 1 presents the demographic characteristics of respondents. The majority of respondents were female (91.3%), had a bachelor degree in nursing (60.0%), and were in staff nurse positions (88.1%). Their ages ranged from 23 to 67 years (
M = 43.6, SD = 11.7). The mean length of experience in the profession and in the organization was 16.6 years (SD = 12.1) and 11.1 years (SD = 9.4), respectively. Respondents worked in a variety of specialty areas, where the greater proportion of them worked in medical units (28.1%), followed by the emergency departments (18.8%) and outpatient clinics (18.1%). Almost half of the respondents (40.5%) reported having received some cultural training.
Ethnicity is an important variable in this study because of its focus on cultural difference and competence. The majority of respondents identified themselves as Caucasian (74.2%) with the two next largest groups being South Asian (12.6%) and Chinese (5.7%). Other ethnic identities included Far East Asian (1.9%), other Asian (1.3%), and First Nation (0.6%). Respondents identified 24 nations as their country of birth, including Canada, the Philippines, Taiwan, Hong Kong, South Korea, Slovakia, China, Ireland, Germany, Netherlands, Paraguay, Hungary, the United Kingdom, India, Russia, the United States, Zimbabwe, South Africa, Ukraine, Australia, Japan, Iran, and Jamaica. These countries were grouped into three major categories: Anglo-Saxon countries, European countries, and Asian countries so that the sub-sample size would be sufficient to support correlation analysis.
3.2. Principal Component Analysis Results
Exploratory factor analysis is used to explore and identify the factor structure of a large set of variables (items), in contrast to confirmatory factor analysis in which the researcher tests specific hypotheses about factor structure. Exploratory factor analysis, PCA in particular, is appropriate during the early stages of instrument development and testing, especially in situations where the goal is to develop a shorter scale [
27,
28,
29].
Prior to performing PCA on the 84 items (
N = 170), we assessed the suitability of the data set for this procedure. The issue of sample size in PCA has been studied and debated for several years. Some authors have argued for a cases-to-variables ratio, but others have demonstrated that overall sample size matters more, in combination with the size of factor loadings and communalities [
30]. For example, one study [
31] showed that sample size was irrelevant as long as there were at least four loadings greater than 0.60 on each factor, or 10 loadings greater than 0.40 with a minimum sample size of 150. Another study [
32] suggested that even a small sample of less than 100 can be reliable if all communalities are 0.60 or greater. Thus, the adequacy of our sample of 170 could only be fully assessed after completing the PCA. Another means of assessing the suitability of the data is through visual inspection of the correlational matrix with the ideal range for correlation coefficients being between 0.30 and 0.70 [
30]. Finally, the Kaiser-Meyer-Olkin Measure of Sampling Adequacy (0.66) and Bartlett’s Test of Sphericity (
p < 0.001) both supported the factorability of the data and suggested that PCA was an appropriate method for performing factor extraction.
Factor extraction refers to the process of grouping a large number of items into a fewer number of factors in order to identify the underlying dimensions. The most commonly used methods are PCA and principal axis factor analysis (PAF). PCA estimates linear components within the data and contributions of a particular item to that component, whereas PAF estimate the items to be linear combination of the unique factors [
30]. Although these two methods differ slightly, they often lead to a similar result in terms of the underlying dimensions [
29]. In this study, we used PCA as the simpler method for variable/item reduction [
28,
33].
Three techniques were used in determining the number of factors to be extracted: Kaiser’s criterion, a scree test, and parallel analysis. Kaiser’s criterion calls for retaining factors that have eigenvalues equal to or greater than 1.0 for further investigation. Although this technique is commonly used, it can be problematic and has been widely criticized in the literature for retaining a large number of factors. This technique is also most appropriate when the case-to-item ratio is 10 to 1 [
34]. Using Kaiser’s criterion in this study yielded 24 factors with eigenvalues of 1.0 or greater; thus, we also used Catell’s scree test, which plots the eigenvalues against factor numbers. We were looking for a sharp break in the slopes and a levelling off, which would indicate “the number of meaningful factors, different from random error” [
35]. Inspection of the scree plot indicated that there was a break after the fourth component with the curve changing direction; however, it was not a clear and sharp elbow-like break which posed some uncertainty about the number of factors that should be retained. Therefore, although the number of factors (4) was congruent with the theoretical foundation of the CCCS, the data were subjected to further investigation using a parallel analysis technique.
In parallel analysis, we used the eigenvalue Monte Carlo simulation technique, in which the size of the observed eigenvalues are compared with those “obtained from a randomly generated data set of the same size” [
36] (p. 183). The number of factors with observed eigenvalues greater than that obtained in the simulated data set determines the number of factors that should be retained. However, critics have noted that a problem with this technique is the possibility of retaining factors that may poorly defined [
37]. The use of parallel analysis in this study suggested a nine-factor solution.
Given the wide range in the number of factors that might best describe the dimensionality of the data, we conducted a series of PCA with a forced number of factors ranging between 4 and 9, comparing the eigenvalues, factor loadings, and communalities of the factor structure. The best solution was obtained with a four-factor solution that accounted for 33% of variance of the full set of 84 items, with the first two factors accounting for 14% and 9% of the variance, respectively. We then employed an orthogonal rotation method (Varimax) that assumes that underlying factors are only weakly correlated, consistent with recommendations in the literature for initial investigations of factor structure [
38].
The next step was to refine the CCCS by deleting 22 items with low factors loadings (i.e., less than 0.40) and low communalities (i.e., less than 0.30) [
35,
39]. There were 12 other items with communalities less than 0.30 but we retained these due to their theoretical relevance and factor loadings of greater than 0.40. After re-running the PCA with the reduced set of items (
k = 62), we found that most (but not all) of the items showed a moderate to strong loading on a theoretically appropriate factor. We then deleted an additional 11 items because they were irrelevant to the factor (three items), cross-loaded on two factors (two items), or cross-loaded on two factors and were irrelevant to the dominant factor (six items). The remaining items were further evaluated for redundancy (i.e., items that were worded similarly or tapped closely into the same idea) and we retained the item with the higher factor loading. For example, we kept the first of two items (“I feel safe in expressing my concerns to management” and “I feel supported in expressing my concerns to management”) because of its higher loading. The deleted items came from across the four factors. The final solution consisted of 43 items that loaded cleanly across four factors explaining 42.3% of the variance. The final analysis yielded a scree plot that indicated the clear presence of four factors.
Table 2 presents the factor loadings for each of the items, their communalities, and the percent of variance explained by each factor.
To interpret and label the four factors, we evaluated the theoretical meaning of the set of items that loaded on each factor. The set of items loading on Factor 1 pertained to the individual’s perception of racialization, assessment of their organization, and personal confidence; therefore, this factor was labelled as “Critical Empowerment.” The items that loaded on Factor 2 reflected self-awareness of the respondent’s own attitudes and values, and recognition of the determinants of power and consequences of both power and cultural differences, so this factor was labelled “Critical Awareness.” Factor 3 items were concerned with the skills needed for safe and competent cross-cultural interactions, so this factor was labelled as “Critical Skills.” Factor 4, labelled as “Critical Knowledge,” pertained to conceptualizations of culture. Thus, the observed factor structure of the refined CCCS closely reflects the original conceptual model of CCC developed by Almutairi, Dahinten and Rodney [
22]. The four components of the CCC model, with associated constituent facets and items, are presented in
Table 3.
Finally, to check our assumptions about the appropriateness of orthogonal rotation, we re-ran the PCA using oblique rotation (Oblimin). The resulting component correlation matrix showing correlations ranging between −0.13 and 0.22 supported our assumptions about the independence of the factors, and the use of Varimax rotation.