and in the final step the model was refined through introduction of additional

and in the final step, the model was refined through introduction of additional relations be- tween manifest and latent variables. The successive testing and modification of dif- ferent models on the same data of course makes it impossible to argue that the goodness-of-fit test for the final model can be evaluated against the w 2 - distribution. This, along with the fact that viola- tions of the assumption of multivariate normality causes the w 2 -statistic to be inﬂated, suggests that the w 2 -statistic should primarily be used for testing differences in fit between different models. The final CFA model, which was used as the baseline model against which different structural models were compared, had a w 2 -value of 1,093.4, df ¼ 598 ( P ¼ 0.00). The RMSEA index was 0.043 for this model, with a 90% confidence interval between 0.043 and 0.047, which indicates acceptable fit of the model to data. Other indices of fit also showed fit to be acceptable (AGFI ¼ 0.86, CFI ¼ 0.92, NFI ¼ 0.85, NNFI ¼ 0.91). Out of the 57 items in the instrument, 38 were included in the final model, which comprised seven latent variables. The reason why the final model included seven factors rather than the hypothe- sized five factors is that the items in the motivation category were found to load on three separate variables: intrinsic motivation , extrinsic motivation, and evaluation and reﬂection . The other four hy- pothesized factors, which were labelled information sharing , learning culture, networking, and creative climate, related to the manifest variables largely as expected. The model also included covariances among residuals of eight pairs of manifest vari- ables that represent minor sources of item overlap. Standardized factor loadings, along with descrip- tive item data, are shown in Table 1. Eight items measured the information sharing factor; all dealt with accessing and sharing scientific information. Three items with the highest loadings (0.69–0.63) asked about the extent to which scien- tific information is shared between and within projects and between R&D sites. A fairly high loading (0.50) was obtained for an item (IS5) that asked about information sharing within projects in the form of reports. Items that asked about fre- quency of access and ease of access to scientific information had lower loadings (0.24–0.50). The learning culture factor was related to nine items, most of which had high loadings on the factor. The highest loadings (0.69) were observed for one item that asked if the organization is good at rewarding new ideas and another item that asked if the company’s stated vision is reﬂected in its actions. Loadings around 0.6 were observed, among other things, for items that asked about effective use of expertise and previous experience and orientation toward creating new ideas.