GESIS Training Courses

Scientific Coordination

André Ernst

Administrative Coordination

Janina Götsche

Assessing the comparability of multiple-indicator constructs with multiple-group confirmatory factor analysis

Köln / Unter Sachsenhausen 6-8
General Topics:
Course Level:
Software used:
Students: 200 €
Academics: 300 €
Commercial: 600 €
Additional links
Lecturer(s): Dr. Daniel Seddig

About the lecturer - Dr. Daniel Seddig

Course description

Many phenomena in social science are not directly observable, such as beliefs, attitudes,  values, life-satisfaction, or well-being. Instead, they are conceptualized as latent variables and indirectly measured by multiple observed indicators that are assumed to reflect a theoretical construct. The connection between a latent variable and its indicators can be formalized as a measurement model and tested with confirmatory factor analysis (CFA) (e.g., Brown, 2015).
In comparative social research, regression coefficients among latent variables or mean scores are often compared across groups (e.g., cultures, countries, companies). However, valid comparisons require that the parameters of the measurement model (e.g., factor loadings, indicator intercepts/thresholds) are equivalent across groups (e.g., Vandenberg & Lance, 2000). Otherwise, comparisons can be misleading, for example, when seemingly real substantive differences in a latent construct are actually due to measurement differences (e.g., Chen, 2008). Researchers can test whether a measurement model is invariant across groups using multiple-group CFA.
The course provides insight to the procedures of testing measurement invariance using the R package lavaan (Rosseel, 2012) and data from the European Social Survey (ESS) on human values and perceived threat due to immigration (Davidov et al., 2020). Optionally, and as a supplement to the methods of testing exact invariance, the alignment optimization procedure (Asparouhov & Muthen, 2014) will be illustrated as an example of approximate invariance.

Target group

Social scientists engaged in (quantitative) comparative (or longitudinal) research.

Learning objectives

The participants should learn to understand literature and analyzes based on structural equation models for comparative data. In addition, they should be able to translate their own research questions on the comparability of latent constructs into statistical models and to analyze them and interpret results using the lavaan/R software.


Participants should have background knowledge in OLS regression analysis. Moreover, some basic knowledge in path analysis/CFA/SEM and using the R program would be helpful.


Recommended readings

More Information