Developing a framework for the management of digital adoption

As the pace and intensity of innovation and organizational change steadily increase due to the digital transformation itself (Brynjolfsson & McAfee, Reference Brynjolfsson and McAfee2011; Kane, Phillips, Copulsky, & Andrus, Reference Kane, Phillips, Copulsky and Andrus2019; Kotter, Reference Kotter2012), the successful implementation of digital technologies requires continuous adoption from employees (Andriole, Cox, & Khin, Reference Andriole, Cox and Khin2018; Cavalcanti, Oliveira, & de Oliveira Santini, Reference Cavalcanti, Oliveira and de Oliveira Santini2022). However, in many cases, the importance of employees’ reactions during digital transformations is underestimated (Kane et al., Reference Kane, Phillips, Copulsky and Andrus2019; Kohnke, Reference Kohnke, Oswald and Kleinemeier2017). Therefore, organizations must recognize the relevance of the psychological processes and better understand how these are triggered by specific management interventions in digital transformation projects. Despite their high practical relevance, few attempts have been undertaken to systematically connect management interventions with technology adoption and its underlying psychological processes (Abbasi, Tarhini, Hassouna, & Shah, Reference Abbasi, Tarhini, Hassouna and Shah2015; Kohnke, Wolf, & Mueller, Reference Kohnke, Wolf and Mueller2011). However, the broader change management literature as a parallel research stream describes a plethora of management factors for organizational change processes, which also include changes relating to digital transformations (Oreg, Vakola, & Armenakis, Reference Oreg, Vakola and Armenakis2011; Rafferty, Jimmieson, & Armenakis, Reference Rafferty, Jimmieson and Armenakis2013). In an effort to systematically integrate research in the general domain of organizational change, Armenakis and colleagues developed a taxonomy of change management factors that distinguished four major categories assumed to impact reactions to change processes (Armenakis & Bedeian, Reference Armenakis and Bedeian1999; Holt, Armenakis, Feild, & Harris, Reference Holt, Armenakis, Feild and Harris2007; Walker, Armenakis, & Bernerth, Reference Walker, Armenakis and Bernerth2007). These four categories are described as content, process, context, and individual factors. The individual factors category (e.g., affect, status, age, gender, or organizational commitment), however, constitute pre-change conditions (Oreg, Vakola, & Armenakis, Reference Oreg, Vakola and Armenakis2011) that are, by definition, often constant and therefore of limited use for managing specific digital transformation projects. The other three categories from the taxonomy provide a great foundation for the categorization and identification of relevant management factors in the specific context of digital transformation processes.

Content factors refer to what is being changed and comprise perceptions of the utility, necessity, and consequences of the change process. In the context of digital transformations, these factors are mainly related to specific characteristics of the digital technology at hand. Key aspects of IT systems frequently found in the literature are the quality of data, functionality provided by the system, and system performance (e.g., DeLone & McLean, Reference DeLone and McLean1992; Liu & Ma, Reference Liu and Ma2006; Wixom & Todd, Reference Wixom and Todd2005). Management may influence these characteristics indirectly by choosing between technologies that differ in these regards and initiating improvements in case quality expectations are not met. Process factors refer to how things are being changed and encompass classical change management factors such as user information and communication (e.g., Amoako-Gyampah & Salam, Reference Amoako-Gyampah and Salam2004; Kohnke, Wolf, & Mueller, Reference Kohnke, Wolf and Mueller2011), user training (e.g., Amoako-Gyampah & Salam, Reference Amoako-Gyampah and Salam2004; Kohnke, Wolf, & Mueller, Reference Kohnke, Wolf and Mueller2011), user participation (e.g., Hartwick & Barki, Reference Hartwick and Barki1994; Rouibah, Hamdy, & Al‐Enezi, Reference Rouibah, Hamdy and Al‐Enezi2009), and user support (e.g., Igbaria, Zinatelli, Cragg, & Cavaye, Reference Igbaria, Zinatelli, Cragg and Cavaye1997; Karahanna & Limayem, Reference Karahanna and Limayem2000). Since process factors can vary widely, in this study, we focus on user information about the transformation as basic process factor needed in any change process (Armenakis & Harris, Reference Armenakis and Harris2002). Context factors refer to the organizational environment in which the change occurs. These factors include organizational conditions that facilitate the implementation process, such as top management support (e.g., Kohnke, Wolf, & Mueller, Reference Kohnke, Wolf and Mueller2011; Rouibah, Hamdy, & Al‐Enezi, Reference Rouibah, Hamdy and Al‐Enezi2009), supervisor support (e.g., Schepers, Wetzels, & de Ruyter, Reference Schepers, Wetzels and de Ruyter2005; Schillewaert, Ahearne, Frambach, & Moenaert, Reference Schillewaert, Ahearne, Frambach and Moenaert2005), or organizational antecedents like the change history (e.g., Hanelt et al., Reference Hanelt, Bohnsack, Marz and Antunes Marante2021). While those antecedents are not malleable in the management of a given change, leaders’ behavior generally is and effectively shapes the change context (Oreg & Berson, Reference Oreg and Berson2019). Therefore, top management and supervisor support are considered as context factors in this study.

Content, process, and context factors are furthermore assumed to influence employee behavior through the psychological processes that they elicit (Oreg & Berson, Reference Oreg and Berson2019; Oreg, Vakola, & Armenakis, Reference Oreg, Vakola and Armenakis2011). Hence, a comprehensive approach to the management of digital transformation should reflect central variables from all three categories and relate them to changes in the psychological processes of digital adoption and actual behavioral change. Indeed, research combining management factors with specific psychological reactions of change recipients has demonstrated its value for understanding how change reactions can be influenced by change management (Jimmieson, Peach, & White, Reference Jimmieson, Peach and White2008; Straatmann et al., Reference Straatmann, Kohnke, Hattrup and Mueller2016).

One of the most cited models to define psychological variables that explain and predict user behavior and adoption of technologies by individuals is the unified theory of acceptance and use of technology (UTAUT) by Venkatesh, Morris, Davis, and Davis (Reference Venkatesh, Morris, Davis and Davis2003). The UTAUT is integrating established mechanisms of previous theories of technology acceptance and behavior in general (e.g., the Theory of Planned Behavior; Ajzen, Reference Ajzen1991; or the Technology Acceptance Model; Davis, Reference Davis1989; Davis, Bagozzi, & Warshaw, Reference Davis, Bagozzi and Warshaw1989) and has been extensively used by many researchers (Blut et al., Reference Blut, Chong, Tsigna and Venkatesh2022; Dwivedi, Rana, Jeyaraj, Clement, & Williams, Reference Dwivedi, Rana, Jeyaraj, Clement and Williams2019; Venkatesh, Thong, & Xu, Reference Venkatesh, Thong and Xu2016). As such, the core elements identified in the UTAUT provide a useful theoretical foundation to describe the psychological processes that can help to explain how management interventions relate to actual behavior. In essence, the UTAUT specifies that the intention to use a system and facilitating conditions are core determinants of actual user behavior. Behavioral intention, in turn, is determined by performance expectancy, effort expectancy, and social influence. Performance expectancy is defined by Venkatesh et al. as ‘the degree to which an individual believes that using the system will help him or her to attain gains in job performance’ (Reference Venkatesh, Morris, Davis and Davis2003, p. 447). Effort expectancy refers to ‘the degree of ease associated with the use of the system’ (Reference Venkatesh, Morris, Davis and Davis2003, p. 450). Social influence describes ‘the degree to which an individual perceives that important others believe he or she should use the new system’ (Reference Venkatesh, Morris, Davis and Davis2003, p. 451). Facilitating conditions refer to ‘the degree to which an individual believes that an organizational and technical infrastructure exists to support the use of the system’ (Venkatesh et al., Reference Venkatesh, Morris, Davis and Davis2003, p. 453) and from a psychological perspective reflect the belief of having the necessary resources and knowledge to use the system (Ajzen, Reference Ajzen1991; Taylor & Todd, Reference Taylor and Todd1995). In addition to its direct influence on actual user behavior, facilitating conditions were considered as a fourth determinant of behavioral intentions later (Blut et al., Reference Blut, Chong, Tsigna and Venkatesh2022; Dwivedi et al., Reference Dwivedi, Rana, Jeyaraj, Clement and Williams2019). Similar to other scholars (e.g., Dwivedi et al., Reference Dwivedi, Rana, Jeyaraj, Clement and Williams2019), we focus on these core concepts of the UTAUT as this study aims to understand how psychological processes mediate the influence of management factors on user adoption.

Based on previous research in the context of technology adoption, we further include three modifications in terms of relationships among the core concepts of the UTAUT. First, it is assumed that social influence may also have an effect on user’s perception of a system to be useful, i.e., performance expectancy (Schepers & Wetzels, Reference Schepers and Wetzels2007; Venkatesh & Davis, Reference Venkatesh and Davis2000). Second, previous research indicated that the latter also is higher, when users perceive the system as being easy to use (i.e., effort expectancy), as ease of use reinforces the expectation of high performance (e.g., Davis, Bagozzi, & Warshaw, Reference Davis, Bagozzi and Warshaw1989; Oliveira, Thomas, Baptista, & Campos, Reference Oliveira, Thomas, Baptista and Campos2016; Venkatesh & Bala, Reference Venkatesh and Bala2008). Third, studies demonstrated that an individual’s perception of having the resources to use a system (i.e., facilitating conditions) is positively related to the perception of a system as being easy to use (i.e., effort expectancy; e.g., Venkatesh, Reference Venkatesh2000; Venkatesh & Bala, Reference Venkatesh and Bala2008; Yi, Jackson, Park, & Probst, Reference Yi, Jackson, Park and Probst2006).

In summary, we build on an established taxonomy of change management factors with a common and slightly modified model of technology adoption to derive an integrative model, which can provide insights into the psychological processes through which management factors exert their impact on employees’ behavior in the case of a large-scale digital transformation project. More specific knowledge about the effect of management factors (i.e., content, process, and context factors) on psychological factors (i.e., performance expectancy, effort expectancy, social influence, and facilitating conditions) provides guidance for the management to identify how favorable user reactions and behavior can be promoted most effectively in digital transformations. Based on the theoretical framework, we adopt a process perspective on the digital transformation project and derive and test specific hypotheses in the following.

A process-oriented perspective on the management of digital adoption

At a fundamental level of digital transformation, often an existing IT system or infrastructure is replaced by a new system, implying a change process from the legacy to a new system. Also from a theoretical perspective, many Information Systems-related theories are rooted in the assumption that variables and their longitudinal relationships change over time, which makes the analysis of the change dynamics a crucial research question (Zheng, Pavlou, & Gu, Reference Zheng, Pavlou and Gu2014). Even so, research in the Information Systems context that follows a process-oriented research perspective is rare. Noteworthy exceptions are the studies provided by Bala and colleagues (Bala & Venkatesh, Reference Bala and Venkatesh2013; Bala, Venkatesh, Ganster, & Rai, Reference Bala, Venkatesh, Ganster and Rai2021) who explicitly integrated temporal components into their longitudinal research and examined how perceptions of employees’ job characteristics (Bala & Venkatesh, Reference Bala and Venkatesh2013) and interpersonal relationships (Bala et al., Reference Bala, Venkatesh, Ganster and Rai2021) change over time during an enterprise system implementation. Investigating the longitudinal trajectory of employees’ perceptions of IT presents an important path for identifying relevant changes in these perceptions and their causes. Thus, the explicit formulation of process-oriented hypotheses for variables and their relationships over time offers deeper insights into specific change dynamics during digital transformations and can strengthen the empirical evidence for often assumed cause-and-effect relationships (e.g., Langley et al., Reference Langley, Smallman, Tsoukas and Van de Ven2013; Venkatesh et al., Reference Venkatesh, Sykes, Aljafari and Poole2021; Zheng, Pavlou, & Gu, Reference Zheng, Pavlou and Gu2014). Therefore, we take on a process perspective and hypothesize relationships between perceived changes in management factors, changes in psychological factors, and changes in employee behavior. To be precise, we will examine the (interindividual) relationships of intraindividual changes in employees’ perception of management and psychological factors as well as in their behavior.

In particular and building on our theoretical framework, perceived technology improvements (content factors) should lead to intraindividual increases in the psychological factors effort expectancy, performance expectancy, facilitating conditions, and social influence – as others are also affected by the changes. Similarly, intraindividual increases in the perception of process factors (e.g., the amount of helpful user information) and in the perception of a supportive context during digital transformations should lead to an intraindividual increase in effort expectancy, performance expectancy, social influence, and facilitating conditions as psychological reactions. As we selected typical management factors that previous research had shown to facilitate user adoption of information systems, all relationships are assumed to be positive (e.g., Ain et al., Reference Ain, Vaia, DeLone and Waheed2019; Kohnke, Wolf, & Mueller, Reference Kohnke, Wolf and Mueller2011). Hence, we hypothesize the following:

In line with the theoretical framework described above, we expect the psychological factors to be related to employee behavior (e.g., Davis, Reference Davis1989; Lee & Wan, Reference Lee and Wan2010; Sun, Bhattacherjee, & Ma, Reference Sun, Bhattacherjee and Ma2009). From a process perspective, we assume usage behavior to change dependent on changing psychological factors. For example, employees might be reluctant to use the current system because it is not seen as particularly useful but rather complicated to use. If the new system is perceived as being more useful and easy to use, the employee may accept and use it even more. Specifically, intraindividual increases in one or more of the psychological model variables are assumed to lead to intraindividual increases in an individual’s intention to use a digital technology and also to an increase in actual user behavior.

Finally, we test the aforementioned assumption that management factors impact employee behavior through the psychological processes elicited (Oreg & Berson, Reference Oreg and Berson2019; Oreg, Vakola, & Armenakis, Reference Oreg, Vakola and Armenakis2011). Translated into a process hypothesis, we examine whether intraindividual increases in perceptions of management factors lead to behavioral changes through intraindividual increases in psychological variables.

The four hypotheses are summarized in the framework for the management of digital adoption (MDA) outlined in Figure 1. It postulates that intraindividual changes in one of the management factors should result in intraindividual changes in psychological variables and user behavior over time. Thus, we argue that positive significant relationships between intraindividual changes of our study variables would provide evidence for a user adoption gain effect during a digital transformation process.

Figure 1.

Proposed management framework of digital adoption.

Organizational setting

The setting for this field study is a multi-year digital transformation project that took place in a large international logistics company. To improve its competitive strength, the company investigated possibilities for optimization and simplification of the current company-wide standard software solution (an ERP system). Part of this investigation was an evaluation of various management factors such as the quality of the IT system (i.e., data, functionality, and performance), the quality of user information, and the level of support by supervisors and management from a user perspective. This evaluation indicated a clear need for change, and a baseline (t1) was set before a digital transformation project was initiated.

The digital transformation initiative was intended to overcome quality gaps by delivering globally standardized and harmonized business processes enabled by a new, user-friendly IT system. The project concentrated on a wide range of business processes covering financial (e.g., cash management, general ledger, or travel expense management), procurement (e.g., invoice management or terminal expenses), and reporting processes. These changes particularly impacted central and globally distributed departments for finance, operations, and services. Changes reflected in the new IT system included the elimination of manual tasks through automation (e.g., electronic invoicing and monitoring risks) and the simplification of tasks to reduce errors and delays in the process flow (e.g., removal of unnecessary data entry fields or approval steps, avoidance of data double-entry, and manual data extraction). Moreover, a new user interface and system functionality (e.g., for reporting and analysis tasks), as well as new roles and accountabilities for employees (e.g., centralizing approval processes and re-defining access rights to the system), were provided. The digital transformation project was facilitated by an organizational change management team. Key objectives of this team were an increase in the current level of management support (i.e., supervisors and senior management) for the IT system and an improvement of the existing communication content delivered through various channels. One previously used channel was the corporate portal, which gave access to a variety of communication assets such as quick reference guides, standard operation procedures, and learning material.

Three and a half years later, after the digital transformation project was completed, a second evaluation was conducted (t2) to assess the impact of changes in management and psychological factors as perceived by the users. At t2, all process, system, and organizational changes were fully implemented.

Survey administration

The surveys were provided to users in English as web-based questionnaires using an online tool. Invited users could access the questionnaire via a single-sign-on procedure to ensure the participation of only eligible users. Organizational data, such as region, country, and location, as well as objective usage data, had been provided by the company and was matched to the survey data by a unique identifier. Both surveys were hosted and conducted by an external research institute. One week before the survey, an e-mail message describing the purpose and content of the survey was sent by the management sponsor to all system users. The invitation e-mail indicated that participation in the survey was voluntary and anonymous. In countries where surveys require co-determination (e.g., works council), the relevant bodies had been involved. During the survey, if the participant left an item blank, the missing value was accepted by the survey tool without any further queries. If completion was interrupted (e.g., the connection was terminated prior to submission for any reason), the survey could be resumed by logging in again via the single-sign-on procedure. The data-gathering period was 3 weeks for each survey, with reminders sent to all invited users after the first and second week.

Sample

At both measurement points, all employees with a user role within the IT system were selected for the study. At t1, 4,638 of the invited 9,886 employees took part in the survey, resulting in a response rate of 46.9%. At t2, 18,094 users were invited and 4,956 participated in the survey, resulting in a response rate of 27.4%. The larger number of system users at the second measurement time resulted from a broader roll-out of the extended functionality. In total, N = 1,095 of the users reported their experience with the IT system at both measurement points. Sample characteristics such as business function or management level indicated that all employee groups affected by the digital transformation project were represented in the sample.

Measures

All measures used for the study were part of the surveys at both measurement points. While the survey included more variables than simply the ones necessary to the study, only those relating to the MDA framework were eventually analyzed. Measures that covered other content (e.g., open comments), that were only relevant for a specific subsample (e.g., filter questions), or that were changed from t1 to t2 were not included in the analysis. Generally, we adapted the items used to capture the theoretical constructs from prior research, with changes in wording to fit the target context and an additional summary item for each construct (see Table 1; the complete questionnaire is available upon request from the authors). We have not collected sociodemographic data in accordance with the policies of the organization’s works council. Participants used a five-point Likert-type scale (1 = strongly disagree and 5 = strongly agree) to respond to the items. All constructs showed high reliability, with Cronbach’s alpha ranging from .85 to .97.

Table 1.

Items used for the scales with corresponding reliabilities across times of measurement

Self-reported system usage served as a proxy to capture actual system use. We adapted two items from Davis (Reference Davis1993) to assess the frequency and intensity of use. Frequency of use was measured on a 5-point scale ranging from 1 = “Use less than once a week” to 5 = “Use several times a day”. Intensity of use was assessed with an open question regarding the number of hours per week spent working with the system. Since the open format led to a large amount of missing and unrealistic values, this item was eliminated from further analysis. To obtain an additional, objective measure of actual user behavior, the number of dialogue steps in the IT system at t2 was collected. The measured count of dialogue steps refers to the user interaction with the system, such as navigating through a menu or searching in a database. This objective measure indicated the intensity of system use by a single user. As it had a positively skewed distribution, we used its logarithm for further analysis. Specifically, the mean logarithmized number of dialogue steps per day of the four months preceding t2 was used to validate the self-report measure obtained in the survey.

Data preparation and preliminary analyses

Across the model variables, 6.6% of the values were missing. Item-level missingness ranged from 0.1% to 17.1% with a tendency of higher values for items positioned toward the end of the survey. As Littles MCAR tests were significant, data were not missing completely at random, χ²(27,836, N = 1,095) = 29,631.23, p < .001. Following the assumption that most missing data can be considered random to some extent (Graham, Reference Graham2009) and thus predictable by other variables in the dataset, we imputed missing data using the widely used EM algorithm (Lin & Tsai, Reference Lin and Tsai2020) implemented in SPSS.

Measurement models

As a prerequisite for examining latent change, strong measurement invariance across time is necessary (Geiser et al., Reference Geiser, Keller, Lockhart, Eid, Cole and Koch2015). We assessed longitudinal measurement invariance by comparing four sequentially restricted measurement models (Widaman, Ferrer, & Conger, Reference Widaman, Ferrer and Conger2010). The fit of the four models was compared assessing differences in the comparative fit index (CFI) and the root mean square error of approximation (RMSEA), interpreting a difference of ∆ ≤ .01 as acceptable (Isiordia & Ferrer, Reference Isiordia and Ferrer2018). Following this procedure, strict longitudinal measurement invariance could be established, with ∆CFI = .006 and ∆RMSEA = .001 when comparing the least restricted with the most restricted model. The measurement model – assuming equal loadings, item intercepts, and item variances across time – shows an acceptable fit with χ ²(5,525, N = 1,095) = 13,602.45 and p < .001, χ ²/df = 2.46, CFI = .923, Tucker–Lewis Index (TLI) = .920, standardized root mean square residual (SRMR) = .070 and RMSEA = .042.

To further test the hypothesized measurement model and discriminant validity, we compared competing models to the original model in which highly correlated latent factors (r ≥ .65) were modeled as one. To compare the models, the Akaike information criterion was examined additionally, with lower values representing better model fit (Schermelleh-Engel, Moosbrugger, & Müller, Reference Schermelleh-Engel, Moosbrugger and Müller2003). In favor of the original model, no other model showed a better fit, with ∆CFI ≥ .009, ∆RMSEA ≥ .004, and the lowest Akaike information criterion for the hypothesized model. Because all constructs were assessed in one survey at both measurement times, we furthermore tested for a potential bias due to the common method of assessment (Lance, Dawson, Birkelbach, & Hoffman, Reference Lance, Dawson, Birkelbach and Hoffman2010; Podsakoff, MacKenzie, & Podsakoff, Reference Podsakoff, MacKenzie and Podsakoff2012). In particular, we tested a single-factor model (Harman’s one factor test; Podsakoff & Organ, Reference Podsakoff and Organ1986) and a model with two additional, orthogonal latent factors representing the measurement method at t1 and t2 (unmeasured latent factor technique, e.g., Jordan & Troth, Reference Jordan and Troth2020; Podsakoff, MacKenzie, & Podsakoff, Reference Podsakoff, MacKenzie and Podsakoff2012). For the latter, all items were associated with their theoretical construct and equally loading on one of the respective method factors (Jordan & Troth, Reference Jordan and Troth2020). The single-factor model showed inacceptable, worse model fit (∆CFI = .392 and ∆RMSEA = .061), and the model including method factors showed acceptable, but not substantially improved model fit (∆CFI = .005 and ∆RMSEA = .001). Thus, common method bias is not expected to have a serious impact in our study. In addition, we examined the correlation of the self-reported and objective measure of system use at t2 to test convergent validity. With r(1,059) = .67, p < .00, both measures were strongly related to each other. Descriptive statistics and intercorrelations of all model variables are included in Table 2.

Table 2.

Manifest means, standard deviations, and correlations of the variables (N = 1,095)

Note. t1 is the first and t2 the second measurement point. In the lower triangle, correlations at t1 are displayed, and in the upper triangle, correlations at t2. Correlations across measurement points are shown in the diagonal.

*** p < .001

Statistical analyses

The data were analyzed using R, version 3.6.0 (R Core Team, 2019), and the packages lavaan, version 0.6 (Rosseel, Reference Rosseel2012), and semTools, version 0.5. (Jorgensen, Pornprasertmanit, Schoemann, & Rosseel, Reference Jorgensen, Pornprasertmanit, Schoemann and Rosseel2021). Regarding the impact of management factors on digital adoption and its underlying psychological process, we employed a latent change score (LCS) approach (e.g., McArdle, Reference McArdle2009) to model true intraindividual changes explicitly and to study both interindividual (between persons) and intraindividual (within a person) variability. First, all constructs of the MDA framework were modeled as latent variables indicated by four to six items at each measurement point. Only system use was included as a manifest variable in the model since it was assessed by a single item. Following Henk and Castro-Schilo (Reference Henk and Castro-Schilo2016), superordinate latent variables (∆ variables) representing the true change from t1 to t2 in each variable were modeled (detailed model specification is available upon request from the authors). A baseline model, including only correlating latent change variables, served to examine mean intraindividual changes that ‘inform the direction in which most of the sample is changing’ (Henk & Castro-Schilo, Reference Henk and Castro-Schilo2016, p. 4) and guide the interpretation of the effects of latent change variables (see Table 3). For example, we refer to an increase in system use when the latent change variable has a significant positive mean, to a decrease when it has a significant negative mean, or simply to higher or lower change scores in system use when it does not have a significant mean.

Table 3.

Intercepts, standardized intercepts, and variance of latent change variables within the baseline model

Note. The standardized intercept can be interpreted as effect size.

* p < .05;

** p < .01;

*** p < .001

The structural latent change model further included the hypothesized regression paths between the latent change variables, as well as the direct effects of the changes in management factors on changes in behavioral intention and system use to test Hypothesis 4. To examine the hypothesized mediating effects, we estimated the indirect effects of changes in all management factors on system use using Monte Carlo Bootstrapping. The indirect effects were added up to a total indirect effect of changes in each management factor on changes in system use.

All structural equation models, including the measurement models, were estimated using a robust maximum likelihood estimator with robust standard errors. We assessed model fit with common indices, such as the χ ² to df ratio, SRMR, RMSEA, CFI, and TLI (Schermelleh-Engel, Moosbrugger, & Müller, Reference Schermelleh-Engel, Moosbrugger and Müller2003). An acceptable fit is indicated by a χ ² to df ratio of χ ²/df ≤ 3, SRMR ≤ 0.10, and RMSEA ≤ .08 (Schermelleh-Engel, Moosbrugger, & Müller, Reference Schermelleh-Engel, Moosbrugger and Müller2003). For CFI and TLI, a value of .90 can be considered a lower bound (Jackson, Gillaspy, & Purc-Stephenson, Reference Jackson, Gillaspy and Purc-Stephenson2009). The robust version of model fit indices is reported when applicable.