Page 139 - WHERE WE WORK - Schlegelmilch

Ideas in place
230). Since a literature search did not provide us with a relevant measure for visual or task transparency (to the best of our knowledge), the research team decided to develop a measure for workplace transparency. First, we generated a pool of five items for each subdimension drawing on the concept of transparency (Bernstein, 2017; Castilla, 2015). Example items were: “I can see when colleagues are present” (Visual transparency) or “I am aware of my colleagues’ expertise” (Task transparency). The items were discussed within the research team, who are experts on the topic to ensure that the items adequately reflected each concept. We also discussed the formulation with a research assistant to reduce item complexity and ambiguity that can arise from double-barreled questions (Hinkin, 1995). This led us to modify two items as they were ambiguously formulated. In the end, we used a 10-item scale (Table 4.1). The construct validation process described hereafter indicated that the workplace transparency scale should be treated as an aggregate variable in the analysis.
We conducted a Confirmatory factor analysis (CFA) to check the validity of the newly-developed scale (correlations in Table 4.2). We evaluated the fit based on the chi-square statistic (χ2), the Comparative Fit Index (CFI), Tucker-Lewis index (TLI), Root Mean Square Error of Approximation (RMSEA), Standardized Root Mean Square Residual (SRMR), and AIC. A model was accepted as adequate when the RMSEA was > 0.06 to ≤0.08 (Browne & Cudeck, 1989), an SRMR of < .08 (Hu & Bentler, 1999) and a GFI of at least .90 (Bentler, 1990). First, a CFA with all indicators under one factor (M1) because it was a new scale. However, the model had a poor fit (χ2 (25) = 124.319, p < .001, CFI = 0.901, TLI = 0.858, RMSEA = 0.104 [0.086; 0.123], SRMR = 0.067, AIC = 164.319). Second, we ran a CFA with two covariates VISU and TASK (M2) and it had a slightly fit (χ2 (34) = 105.059, p < .001, CFI = 0.932, TLI = 0.910, RMSEA = 0.075 [0.059, 0.092], SRMR = 0.051, AIC = 147.059). Third, the results showed that the model with a second-order transparency latent variable (with covariates VISU and TASK; M3) had good fit (χ2 (33) = 88.026, p < .001, CFI = 0.948, TLI = 0.928, RMSEA = 0.067 [0.051, 0.085], SRMR = 0.047, AIC = 132.026), with the factor loadings being between .51 and .75, with one exception of one item (VISU1)
137