*3.3. Independent Variables*

To measure the internal stakeholder change beliefs, we adopted the 24-item Organizational Change Recipients' Beliefs Scale (OCRBS) developed by Armenakis et al. (2007) with slightly altered wording to fit this study (e.g., "There is a need for the university to improve its operations in teaching, research, and community engagement" (discrepancy), "The project outputs are relevant to the current situation of the university" (appropriateness), "My immediate supervisor encourages me to take advantage of project interventions at work" (principal support)). Organizational characteristics were assessed using a 7-item scale developed from institutionalization models by Armenakis et al. (1999) and Cummings and Worley (2009), e.g., "The university's structure and leadership is flexible to enable smooth project implementation", "The university environment is stable, allowing changes to take root." We included other specific characteristics related to the context of the university as an organization (Brown 2012; Patria 2012). Project characteristics were measured using a 6-item scale developed using the Cummings and Worley (2009) model but also including items representing characteristics of development cooperation projects (Gajic and Palcic 2019; Ika and Hodgson 2014; Tekinel 2013) (e.g., "The project interventions have clear goals understood by all stakeholders," "There is clear coordination and sponsorship of the projects at the level of top management"). Tables 1 and 2 present the reliability and validity of the measures for the independent variables.

**Table 1.** Measurement model evaluation results.



**Table 2.** Latent variable correlations (left) and HTMT ratios (right).

<sup>a</sup> Diagonal bold figures are square roots of each construct AVE (Fornell and Larcker 1981); <sup>b</sup> HTMT ratios less than 0.85 (Henseler et al. 2015).

#### *3.4. Dependent Variables*

The internal stakeholder perceptions of the project or change institutionalization process were assessed using a 10-item scale developed from models and studies by Armenakis et al. (1999), Buchanan et al. (2005), and Cummings and Worley (2009). The items covered the actions identified above. For example, "There is a visible commitment to the outcomes of the project from the bottom to the top levels of the university" (explicit commitmentrelated actions), "New university policies have been formulated and adopted to support project interventions" (implicit structural-related actions), and "project interventions are integrated into the strategies and processes of schools, departments, and units" (integrationrelated actions). Tables 1 and 2 show the reliability and validity measures for the dependent variable.

#### *3.5. Analytical Techniques*

We used SPSS (IBM) version 25 to analyze the demographic data and generate descriptive statistics for the variables in our model, and we used SmartPLS 3.3.3 software to determine the statistical relationships between the latent variables in the model. According to Hair et al. (2019) and Sarstedt et al. (2019), PLS-SEM is a reliable analytical tool for complex models and small samples. This was chosen as an appropriate approach for this study, in which we had a small sample drawn from a small population of respondents, and the model consisted of three exogenous variables, one endogenous variable, and a mediator variable. In project management, for example, this analytical technique has been used to investigate the relationship between project management capabilities and project success (Irfan et al. 2019), to determine how project management self-efficacy predicts project performance (Blomquist et al. 2016), to investigate the impact of stakeholder attributes on disaster recovery project performance (Mojtahedi and Oo 2017), etc. We performed PLS-SEM analysis in two steps, first assessing the measurement model by determining indicator loadings, internal consistency reliability, convergent validity, and discriminant validity. The PLS-SEM algorithm calculates item loadings for each latent variable iteratively, with low-loading items eliminated until an acceptable set is reached. Step two involved evaluating the structural model by determining model parameters and direct and indirect effects, and assessing the significance levels of the parameters and relationships using a bootstrapping procedure with 5000 subsamples (Hair et al. 2019).
