Introduction to structural equation modeling Ned Kock SEM

Introduction to structural equation modeling Ned Kock

SEM techniques • Structural equation modeling (SEM) techniques can be: – Covariance-based – e. g. , those employed by the statistical software analysis tool called LISREL. – Variance-based – e. g. , those employed in partial least squares (PLS) analysis. • SEM techniques are known as second generation data analysis techniques. • SEM allows for the modeling and testing of relationships among multiple independent and dependent constructs, all at once.

Constructs, indicators and paths • Construct – This is a theoretical concept that is not directly measurable. Also known as latent variable. • Indicator – Is a measurable variable used to represent a construct (e. g. , item on a questionnaire). Also referred to as manifest variable, item, and indicant. • Path – Is the link between constructs, or from construct to indicator. Also known as link, and often measured through a path coefficient.

Path coefficient between Y and X = standardized partial regression of Y on X controlling for the effect of one (e. g. , Z) or more variables. X Partial regression (standardized) of Y and X, controlling for Z. Y Mathematical formula Z Partial regression (standardized) of Y on Z, controlling for X. Diagrammatic representation

Path coefficient example Note: Hypothetical situation R 0. 10 W Mathematical formula E Note: This is a simple linear regression model, where R, W and E are manifest variables. 0. 36 Diagrammatic representation

Endogenous vs. exogenous • Exogenous construct – This is a construct that is independent of any other constructs. – No other constructs point at it in an SEM diagram. – Also known as exogenous latent variable. • Endogenous construct – This is a construct that depends on one or more other constructs. – Is pointed at by one or more constructs in an SEM diagram. – Also known as endogenous latent variable.

SEM model components Exogenous construct (a. k. a. independent construct) Indicator Construct (a. k. a. latent variable) Path coefficient Endogenous construct (a. k. a. dependent construct) Interaction effect construct (a. k. a. moderating effect construct) Source: Chin (2001)

Reflective measurement • In this form of construct measurement, paths connecting construct to indicators are directed towards the indicators. • The indicators are supposed to load strongly on the construct. • Such constructs are often designated as latent constructs (or reflective latent constructs).

Reflective measurement example • Construct – New product development team effectiveness • Indicators (question-statements answered on a Likert-type scale) 1. The product met or exceeded volume expectations. 2. The product met or exceeded sales dollar expectations. 3. The product overall met or exceeded sales expectations.

Formative measurement • In this form of construct measurement, paths connecting construct to indicators are directed towards the construct. • The indicators are not assumed to have to load strongly on the construct. • Such constructs are often designated as formative latent constructs. • Only variance-based SEM techniques (e. g. , PLS) can deal with formative latent constructs.

Formative measurement example • Construct – Team electronic communication use • Indicators (question-statements answered on a Likert-type scale) 1. 2. 3. 4. 5. 6. 7. The team used e-mail to fellow team members (1 to 1). The team used e-mail to team distribution lists (1 to many). The team used team messaging boards or team discussion forums. The team used shared electronic files. The team used Lotus notes to facilitate sharing information among team members. The team used electronic newsletters that covered project information. The team used auto routing of documents for team member and management approval. 8. The team used file transfer protocols (FTP) to attach documents to e-mails and Web pages. 9. The team used a Web page dedicated to this project. 10. The team used a Web page for this project that contained project specs, market research information, and test results. 11. The team used voice messaging. 12. The team used teleconferencing. 13. The team used video conferencing 14. The team used desktop video conferencing 15. The team used attached audio files to electronic documents. 16. The team used attached video files to electronic documents.

The SEM advantage • The ability to test multiple relationships at once differentiates SEM techniques from several first generation regression techniques such as: – ANOVA. – MANOVA. – LOGIT. – Linear regression. • Generally, first generation techniques allow for the analysis of a significantly more limited number of relationships between independent and dependent variables at once.

SEM techniques usage Notes: • Information & Management (I&M), Information Systems Research (ISR), and Management Information Systems Quarterly (MISQ) are top-tier journals in the field of information systems. • Wynne Chin, one of the developers of PLS-Graph (a widely used PLS-based SEM analysis tool), is an information systems researcher. • SEM tools are also widely used in other disciplines, mostly behavioral science research-based or related disciplines, for the causal modeling of multivariate datasets where complex webs of relationships between variables are tested.

Techniques comparison

Techniques capabilities

Acknowledgements Adapted text, illustrations, and ideas from the following sources were used in the preparation of the preceding set of slides: 1. 2. 3. 4. Gefen, D. , Straub, D. W. , & Boudreau, M-C. (2000). Structural equation modeling and regression: Guidelines for research practice. Communications of the AIS, 4(7), 1 -76. Kock, N. and Lynn, G. (2005), The E-collaboration Paradox: A Study of 290 New Product Development Teams, Proceedings of the 16 th Information Resources Management International Conference, Khosrowpour, M. (Ed), Idea Group Publishing, Hershey, PA, pp. 444 -448. Kock, N. (2015). Warp. PLS 5. 0 User Manual. Laredo, TX: Script. Warp Systems. PLS-Graph User’s Guide, by W. W. Chin. Publisher: Soft Modeling Inc. (2001). Final slide