MIS Research and Practice: Data Analysis with SEM Techniques – Part I

On August 21, 2010, one reader commented on structural equation modeling (SEM) techniques as follow: “I don’t really know enough about SEM, I know some, but I definitely think I need to know more. Where is the best place to start for someone who has a basic understanding, but wants to learn more?”

Dear reader, the purpose of this post is to provide relevant information on Structural Equation Modeling (SEM) to you. I also want to use this opportunity to emphasize that SEM and other second generation data analysis techniques are increasingly being applied in management information systems (MIS) research and practice. I will be posting additional materials on the use, usefulness, and ease of use of SEM in MIS research and practice on my weblog.

SEM techniques are important for MIS research and practice because they provide powerful ways to address key IT research and practice problems such as understanding IT usage, end-users satisfaction, IT strategic alignment, IT project planning and implementation, knowledge management projects, and so on. These techniques can be extremely useful to IT executives who have IT substantive knowledge for modeling, exploration, and interpretation of results.

Structural Equation Modeling

Structural equation modeling (SEM) techniques refer to a range of multivariate methods aimed at examining the underlying relationships (or structure) among multiple predictor and criterion variables in a model. A structural equation modeling process requires two steps: (a) building and testing a measurement model, and (b) building and testing a structural model. The measurement model serves to create a structural model including paths representing the hypothesized associations among the research constructs.

Various scholars compared and contrasted SEM techniques to other multivariate techniques such as multiple regression analysis, multivariate analysis of variance (MANOVA), and canonical correlation analysis (CCA). While multiple regression analysis assesses the relationship between several independent (or predictor) variables and a dependent (or criterion) variable, MANOVA assesses the relationship between two or more dependent variables and classificatory variables or factors. In contrast, SEM techniques represent only a single relationship between the dependent and independent variables.

Multiple regression and MANOVA do not consider the association effects between criterion and predictor variables that generally characterized human and behavioral problems in management. These techniques are unable to examine a series of dependence associations concurrently. In contrast, SEM techniques can expand the statistical efficiency and explanatory ability for model testing with a single comprehensive framework. According to various scholars, SEM techniques have the ability to (a) model relationships among multiple predictor and criterion variables, (b) represent latent variables (or unobserved concepts) in those relationships, and (c) account for measurement error in the estimation process.

Dow et al. (2008) compared three structural equation techniques: path analysis, item level, and parcel. Dow and his colleagues noted that these techniques produced similar structural results even though they differed in the degree they fit the data and the degree of explained variance. Path analysis is considered as an extension of the regression model with a regression performed for each variable in the model as endogenous variable on others indicated by the model as causal. Item level SEM uses individual items as measured indicators for latent variables and requires a large number of parameters to estimate the fit of the model to the data. Unlike path analysis, this method can test hypotheses regarding the structural (path) model and the measurement relations. Parcel SEM used item parcels to conduct analyses with latent variables. Partial sum of responses to individual items were derived from the parcels derived from partial sum of responses to individual items. Our research study, for example, used path analysis to build and test a business-IT strategic model for multinational corporations.

Various vendors propose different SEM software packages: Amos 5, EQS 6, LISREL 8, Mplus 3, PLS Graph 3, PROC CALIS, and Mx. For our study, LISREL 8 was used for statistical analyses. LISREL is a multivariate analytical software package intended for standard and multilevel structural equation modeling techniques. These methods accommodate different data types including complex survey data on continuous variables and simple random sample data on ordinal and continuous variables.

One can note an increased use of structural equation modeling (SEM) techniques in the social sciences and in information systems research. Prior research used these techniques to address business-IT strategic alignment. Kearns and Lederer (2003) used SEM to build and test a model that examined the relationship between business-IT strategic alignment and organizational strategies. Their research model contained 6 constructs, 26 items, general demographics, and 8 hypotheses. Kearns and Sabherwal (2007) used SEM to create a structural model exploring the linkage between IT strategic alignment and IT business value. Their research model contained 9 constructs, 44 items, and 10 hypotheses.

Earlier studies using SEM did not propose models of business-IT strategic alignment for multinational corporations (MNCs). Our research study (Nkoyock, Spiker, Schmidt, & Martin, 2010) used SEM to analyze data collected through the stratified random sampling of two groups of IT and business managers. Our study drew upon SEM techniques to build and test a business-IT alignment model for MNCs with 10 constructs, 53 items, and 11 hypotheses.

SEM Modeling Process

SEM modeling process is an incremental approach specifying the procedures for testing a model. SEM analyses require essentially a 2-step modeling process of building and testing measurement and structural models: (a) measurement model, and (b) structural model.

The measurement model serves to test the reliability and validity of the measures based on the research model. This model postulates the relationship between the measured items and the underlying constructs or factors. The achievement of “best fitting” measurement model consists of attaining a non-significant chi-square statistic and the recommended values of other goodness-of-fit indices. The second step of the SEM modeling process is the test of the structural model.
Our research study used the measurement model to create a structural model, including paths representing the hypothesized associations among research constructs.

The structural model is the path model defining the relationships among the latent or unobserved variables. The structural model specifies which latent variables cause changes in the values of other latent variables. Various scholars posited the achievement of “best fitting” structural model requires the incorporation of theory, substantive knowledge, previous experience, or other guidelines to discern which independent variables predict each dependent variable.

In our research study, the initial structural model included three latent exogenous variables (organizational emphasis on knowledge management, management of perceived environmental uncertainty, and management of transnational IS strategies), seven latent endogenous variables represented the remaining constructs, and the hypothesized direct paths. The main outcome of our study is a theoretical and practical perspective of business-IT strategic alignment for multinational corporations. For more details about the results of this research study, please visit www.nkoyock.net.

Some References:

Dow, K. E., Jackson, C., Wong, J., & Leitch, R. A. (2008). A comparison of structural equation modeling approaches: The case of user acceptance of information systems. Journal of Computer Information Systems, 48(4), 106-114.
Kearns, G. S., & Lederer, A. L. (2003). A resource-based view of strategic IT alignment: How knowledge sharing creates competitive advantage. Decision Sciences, 34(1), 1-29. doi: 10.1111/1540-5915.02289
Kearns, G. S., & Sabherwal, R. (2007). Strategic alignment between business and information technology: A knowledge-based view of behaviors, outcomes, and consequences. Journal of Management Information Systems, 23(3), 129-162.
Nkoyock, A., Spiker B. K., Schmidt T., & Martin, C. (2010). Business-IT strategic alignment for complex multinational corporations: The case of the U.N. Secretariat. MIS Quarterly, to be published soon.