30 chapter 2 theory-based components of transilience reasonably well. However, for each component, a reverse-coded item correlated more strongly with a component they were not assigned to (see Table 2.2). Therefore, we removed these three items from the scale before calculating mean scores. Furthermore, correlations were relatively lower for the persistence component, compared to the other two components, suggesting that the persistence items had lower quality. Next, we tested whether a three-factor model fits the data better than a unidimensional model, using the package lavaan in R for SEM. As expected, the three-factor model fitted the data significantly better than a unidimensional model, χ2 (3) = 29.3, p < .001 (see model fit indices in Supplementary Material), indicating that transilience consists of three distinct components. We further examined content validity by testing whether the transilience scale, though comprising of three components, indeed assesses a single construct. For this, we used two indicators: the Haberman method (Haberman, 2008; Reise et al., 2013), and omega hierarchical (ωh; Revelle, n.d.). The Haberman method is considered a minimal test to establish whether sub-scores in a multidimensional scale have any psychometric justification (Reise et al., 2013). In multidimensional scales with intercorrelated components (as we assume is the case for transilience), the aggregated total score (i.e., transilience) is often a better estimate of the true score on a component (e.g., persistence) than the observed score on the component; in this case, the latter provides no added value to the total score and is therefore recommended neither to report nor interpret it (see Reise et al., 2013 for elaboration). The Haberman method compares the proportional reduction in mean squared error based on total scores (PRMSET) and subscale scores (PRMSES). When PRMSET > PRMSES, the score on a component adds little value to the aggregated total score (Reise et al., 2013). Using the package subscore in R (version 4.0.2), we found that PRMSET > PRMSES for all transilience components (see Table 2.3), suggesting that the total transilience score is what should be reported and interpreted. Omega hierarchical reflects the proportion of variance in a multidimensional instrument that can be attributed to a common factor (Revelle, n.d.). Using the psych package in R (Revelle, 2022), we found ωh = .67, thus 67% of the variance in the transilience scale can be attributed to a common factor.4 The reliability of the resulting overall transilience scale (15 items) was good (see Table 2.4). The mean score on the transilience scale was well above the mid-point scale (see Table 2.4), indicating that, on average, people perceive they can be transilient in the face of climate change risks. 4 Although there are no official guidelines on the interpretation of omega hierarchical, according to Revelle (n.d., p. 228-230) a value of ωh =.48 indicates a large general factor and small group factors. Hence, ωh =.67 indicates that the scale mostly reflects a single, general factor.
RkJQdWJsaXNoZXIy MTk4NDMw