Integrated Generalized Structured Component Analysis: On the Use of Model Fit Criteria in International Management Research.
| Verfasser | Cho, Gyeongcheol |
| Posten | RESEARCH ARTICLE |
1 Introduction
International management (IM) and international business (IB) research is characterized by the complexity of the phenomena it investigates and their relationships (e.g., Eden & Nielsen, 2020). While some of the elements we study can be measured directly (i.e., observed variables), others can be measured only indirectly (i.e., constructs). Structural equation modeling (SEM) is a general multivariate method that can handle the inherent complexity of IM/IB phenomena by specifying and examining the intricate interrelationships between observed variables and constructs (e.g., Hult et al., 2006; Richter et al., 2016). With (common) factors and components, previous SEM research has used two statistical approaches to represent constructs derived from measurement theory. Depending on whether a construct is represented by a factor or a component, SEM has evolved into two domains - factor-based SEM and component-based SEM (e.g., Joreskog & Wold, 1982; Rigdon, 2012; Rigdon and Sarstedt 2021; Tenenhaus, 2008). From a methodological perspective, covariance structure analysis (CSA; Joreskog, 1970) is considered a standard statistical approach in factor-based SEM (Joreskog, 1970). In contrast, generalized structured component analysis (GSCA; Hwang & Takane, 2004, 2014) and partial least squares path modeling (PLSPM; Lohmoller, 1989; Wold, 1982) are the most salient approaches used in component-based SEM (Hwang et al. 2020; Khan et al. 2019).
IM/IB researchers have increasingly realized that the two representations of constructs are theoretically incompatible (e.g., Beugelsdijk et al., 2018; Coltman et al., 2008; Richter et al., 2016) and that it is recommended using factor-based SEM approaches (e.g., CSA) to estimate models with factors while using component-based SEM approaches (e.g., GSCA) to estimate models with components. To ignore the match between construct representation and method would most likely produce biased estimates and model predictions, potentially leading to invalid inferences (Cho et al., 2021; Hair & Sarstedt, 2019; Hwang et al., 2021; Rigdon et al., 2017; Sarstedt et al., 2016). In order to test theoretical models that comprehensively explain IM/IB phenomena, researchers in this field often need to include both factors and components in the same model to analyze the pathways that lead to an outcome. For example, Obadia (2013) developed a component measure for foreignness-induced cognitive disorientation and assessed its nomological validity by estimating its impact on trust and export performance, both of which had been measured using factors. Similarly, Gilbert and Heinecke (2014) analyzed the impact of regional management autonomy and product/service adaptation on multinational corporations' regional success, drawing on a mix of components and factors. Further, IM/IB research is characterized by common control variables conceptualized by both components and factors (Cuervo-Cazurra et al., 2020; Nielsen & Raswant, 2018) and by using primary and secondary data combined, which often requires the joint use of components and factors (e.g., Cerar et al., 2021). IM/IB researchers, who frequently consider components and factors in their models, seem to be unaware of the methodological challenges brought on by accommodating both construct types in a single analysis. In fact, most SEM approaches are ill-equipped for examining factors and components in the same model, and efforts to suitably extend them come with various challenges and limitations (e.g., Grace & Bollen, 2008; Sarstedt et al., 2016). Addressing this concern, Hwang et al. (2021) recently developed integrated generalized structured component analysis (IGSCA), a component-based approach to SEM that accommodates both construct types, while maintaining the flexibility of the original GSCA method. Due to being so recently introduced, IM/IB researchers have not integrated the IGSCA into their methodological toolbox. Nevertheless, it has obvious relevance for the field.
Thus, this paper's first objective is to enhance IM/IB researchers' awareness of how to correctly represent constructs and to introduce IGSCA as a novel analytical approach for handling such models. Importantly, IGSCA not only facilitates handling different kinds of constructs, but also comes with various overall fit measures that enable researchers to evaluate and compare different models. This sets the method apart from the more popular PLSPM method, which is restricted in terms of model fit assessment (Hair et al., 2019; Richter et al., 2016; Sarstedt et al., 2022). Research in the CSA domain has transferred well-known measures, such as the goodness-of-fit index (GFI; Joreskog & Sorbom, 1986) and the standardized root mean square residual (SRMR), into a GSCA context. Cho et al. (2020) recently evaluated the efficacy of GFI and SRMR and derived cut-off values for their use in models with components only (i.e., GSCA). However, these findings and recommendations cannot readily be transferred to IGSCA, which considers a combination of components and factors. In light of the above, this study's second objective is to address this concern by conducting a simulation study in which we examine the measures' efficacy regarding differentiating between correct and incorrectly specified models that include both factors and components. Therefore, we advance understanding of IGSCA by evaluating the usefulness and applicability of two fit indices that are particularly useful in the assessment and comparative evaluation of models with different complexities. Such an assessment is of specific importance to IM/IB because this research area is characterized by the often-simultaneous use of factors and components (Richter et al., 2016), complex structural models (Eden & Nielsen., 2020), and conflicting theoretical frameworks and perspectives (Cuervo-Cazurra et al., 2020).
In pursuing these objectives, our study contributes to IM/IB research in two important ways. First, by calling attention to the correct representation of factors and components, we aim to increase researchers' awareness of the challenges of including components and factors in the same model without adequate analytical approaches. If IM/IB research does not recognize and respond to the often, complex mix of components and factors it could hinder the field's theoretical and practical advancement (Casson, 2018; Eden & Nielsen, 2020). IM/IB researchers who are not aware of the issues potentially arising in models that include both components and factors, might not analyze the data correctly, which could possibly result in biased findings and misleading theoretical and practical implications. In their recent review of methodological shortcomings in IM/IB research, Aguinis et al. (2020) identified deficient measures (i.e., the measure does not provide a complete understanding of the construct) and the multidimensionality of constructs in IM/IB research as problematic, as the analysis can become a significant methodological challenge. As a solution to this specific problem, they have recommended using analytical techniques based on the conceptualization of a construct. In our article, we introduce IGSCA to the IM/IB field as an analytical approach that responds to such recent calls (e.g., Aguinis et al., 2020; Knight et al., 2021).
Our paper's second contribution, not limited to IM/IB research, is the assessment and evaluation of two fit indexes that can be used in IGSCA to indicate whether the proposed model reaches the minimum requirements based on the observed data. In doing so, we address recent calls to increase IM/IB research findings' reproducibility and trustworthiness (e.g., Aguinis et al., 2017; Cuervo-Cazurra et al., 2020; Delios, 2020; Norder et al., 2021). Specifically, the trustworthiness in empirical analysis can be improved by using methodological approaches that assure the IM/IB research findings are based on a rigorous, unbiased analysis (Cuervo-Cazurra et al., 2020) and by transparently reporting what potential limitations and biases the used methods have (Meyer et al., 2017). IGSCA's ability to use fit indexes enables researchers to evaluate theoretical model's possible consistency with the observed data and to compare alternative models, which is an important element of scientific research (Danks et al. 2020; Sharma et al. 2019, 2021). Our results suggest cut-off values for each of the indexes, which minimize Type I and Type II error rates under different conditions. We also illustrate the two indexes' application using a cultural intelligence model, which IM/IB researchers commonly use to assess a person's intercultural competence (e.g., Matsumoto & Hwang, 2013). Our empirical example not only illustrates the indexes' efficacy but also answers recent calls to improve our understanding of the role personality traits and international experience play in shaping cultural intelligence (Michailova & Ott, 2018; Richter et al., 2020).
The remainder of the paper is structured as follows: First, we provide a brief overview of factors and components, giving examples of their use in IM/IB research. We provide a brief review of different SEM approaches' capacity and shortcomings in handling factors and components in the same model and of these SEM approaches' use in IM/IB research in the last decade. Second, we report the results of a simulation study which investigated the use of cut-off values in assessing model fit in IGSCA. Third, we present the results of an empirical application of the fit indexes in an IM/IB context. Finally, we summarize the fit indexes' behaviors, suggest cutoff criteria, and discuss potential issues that researchers should consider when they utilize the indexes, with a particular emphasis on IM/IB studies.
2 On Using Models with Constructs in IM/IB Research
2.1 A Brief Overview and Examples of Factors and Components
SEM is a general multivariate method for specifying and...
Um weiterzulesen
FORDERN SIE IHR PROBEABO ANCOPYRIGHT GALE, Cengage Learning. All rights reserved.
