7. SEM | SPSS AMOS Lecture Series - Building a Basic Measurement Model in AMOS - Research Coach

Structural equation modeling in AMOS splits work into two linked parts: a measurement model that checks whether constructs are measured reliably and validly, and a structural model that tests how constructs relate through paths and their significance. Parameters sit at the center of both stages—each parameter captures the size and direction of a relationship between variables or between a latent construct and its indicators. In AMOS, some parameters are fixed (often to set a reference scale), while others are estimated freely from the data, including factor loadings and error terms at the measurement level and path coefficients at the structural level.

The lecture then walks through building a basic measurement model in IBM SPSS AMOS 28 Graphics using a single construct: servant leadership. Because servant leadership can’t be observed directly, the model treats it as a latent variable measured by seven questionnaire items (seven indicators). The practical goal is to draw the latent construct, connect it to the seven indicators with arrows, and include error terms for the indicators. AMOS requires importing the dataset first, which is done via the “Select Data File” option, followed by checking variables in the dataset through “List Variables.”

With the data loaded, the model is built in two different AMOS workflows. In the first (simpler) workflow, the user selects an option to draw a latent variable with indicators. The latent variable is created, the number of indicators is set to seven, and AMOS automatically generates arrows from the latent construct to the indicators plus associated error terms. The layout is adjusted using tools like “Rotate Indicators of the Latent Variable” and “Move Objects,” ensuring the indicators and error terms stay properly aligned. The latent variable is named “servant leadership,” error terms are named (via plugins), and the model is prepared for estimation by selecting analysis properties such as standardized estimates and modification indices (with additional options deferred to later videos). After saving the file to a writable directory, “Calculate Estimates” produces outputs like factor loadings, which can be displayed in standardized form.

The second workflow builds the same measurement model more manually using “unobserved variables” and then “observed variables” (indicators). This approach requires drawing the latent construct, adding seven observed indicator boxes, drawing paths from the latent variable to each indicator, and then explicitly adding error terms to each indicator. The manual method also demands a key technical fix: one parameter must be fixed as a reference point (for example, setting a regression weight to 1 on one arrow) so the model can be identified and run. Once that constraint is set, the model can be estimated and outputs viewed, with the lecture previewing that future sessions will expand to multiple constructs and measurement models beyond a single latent variable.