Review of Dynamic Structural Equation Models for Real-Time Consumer Behaviour: Methodological Advances and Applications Insights

MAGASI, Chacha

doi:10.2478/mdke-2025-0004

Figures & Tables

Established inclusion criteria
Source: own processing — Established inclusion criteria Source: own processing

A Comparative framework of traditional and dynamic SEM

Aspect	Traditional SEM	Dynamic SEM
Introduction	It focuses on analysing static relationships between latent variables by assessing covariance structures (Westland, 2015).	Dynamic SEM integrates temporal dynamics, modelling the evolution of consumer behaviours over time (Hamaker et al., 2021).
Key Advances and Applications	Explore static theoretical models in consumer preferences and behaviour to offer relatively stable insights over time (Ghasemy et al., 2020).	Pioneering real-time analysis of consumer behaviour, providing immediate feedback and adaptable marketing strategies (Kronemann, 2022).
Methodological Foundations	Evaluate fixed and static theoretical relationships among variables based on both path analysis and factor analysis (Ghasemy et al., 2020).	Utilises dynamic factor models, latent growth models, and state-space models to monitor real-time behavioural changes (McNeish & Hamaker, 2020).
Data Sources	Primarily relies on cross-sectional or longitudinal data, capturing static behavioural snapshots (Iskamto & Gunawan, 2023).	Requires high-frequency, real-time data from digital platforms such as e-commerce and social media (Tao et al., 2022).
Temporal Scope	Primarily retrospective, focusing on historical data or pre-set intervals for analysis (McNeish & Hamaker, 2020).	Offers predictive modelling, forecasting future consumer behaviours based on past and real-time data (Bolton et al., 2018).
Model Flexibility	Rigid model structure; most effective for environments with consistent and predictable behaviour patterns (Ghasemy et al., 2020).	Highly flexible, allowing for the adaptation of models in response to rapidly changing consumer behaviours (Hamaker et al., 2021).
Real-time Data Adaptation	Unable to handle real-time data, limiting its use in fast-paced, evolving environments (Iskamto & Gunawan, 2023).	Built to process and react to real-time data, it is ideal for dynamic, fast-changing markets (Tao et al., 2022).
Time Sensitivity	Best suited for analysing long-term trends; lacks real-time responsiveness (McNeish & Hamaker, 2020).	Highly time-sensitive, capable of providing continuous, real-time analysis and feedback (Kwasnicka et al., 2019).
Research Fit	Ideal for studies focusing on stable, long-term relationships and testing theoretical models with static data (Westland, 2015).	Best for research requiring real-time behaviour tracking, especially in dynamic industries (Bolton et al., 2018).
Software and Tools	Supported by tools like AMOS, LISREL, and EQS, which are suited for traditional, static data analysis (Hu & Lovrich, 2020).	Requires advanced software like Mplus and OpenMx to handle complex, dynamic data sets in real-time (Hu & Lovrich, 2020).
Real-World Applicability	Limited effectiveness in industries like e-commerce and social media, where consumer behaviours change rapidly (Mehedintu & Soava, 2022).	Highly applicable in digital marketing, app development, and other industries where consumer preferences shift quickly (Kronemann, 2022).
Applications in Marketing Research	Commonly used for analysing stable, long-term consumer patterns and testing fixed hypotheses (Uju & Arizal, 2023).	Used extensively in real-time marketing to adjust campaigns dynamically based on evolving consumer behaviours (Kronemann, 2022).
Key Advantages	Well-suited for analysing stable relationships, useful for theoretical model testing and validation (Westland, 2015).	Excels in tracking fast-changing behaviours, providing actionable insights in real-time for adaptive marketing strategies (Hamaker et al., 2021).
Limitations	They cannot easily adapt to rapid behavioural changes and are limited to retrospective analysis (Sharma et al., 2024).	Computationally intensive, requiring sophisticated algorithms and large data sets for optimal performance (Bolton et al., 2018).
Conclusion and Future Directions	Remains valuable for static analysis, though less effective for dynamic, evolving behaviours (Ghasemy et al., 2020).	Expected to lead future research in real-time behaviour analysis, with potential for further advancements in computational techniques (Hamaker et al., 2021).
Recommendations for Future Research	Suggests incorporating dynamic elements to enhance relevance and application in fast-changing environments (Xu et al., 2020).	Calls for further refinement in computational methods and interdisciplinary research to overcome challenges (Kim et al., 2022).

Comparative analysis of performance metrics for traditional SEM and dynamic SEM models

Metric/Index	Traditional SEM	Dynamic SEM	References
Mean Absolute Error (MAE)	Higher MAE values, indicating less accuracy in prediction (e.g., > 0.10)	Lower MAE values (e.g., < 0.05), reflecting better accuracy in capturing changes	Browne & Cudeck (1992); Zhu, Raquel, & Aryadoust (2020)
Root Mean Squared Error (RMSE)	Higher RMSE (e.g., > 0.10), may not effectively capture short-term changes.	Lower RMSE (e.g., < 0.05), particularly effective in predicting short-term fluctuations	Browne & Cudeck (1992); Zhu, Raquel, & Aryadoust (2020)
R-squared (R²)	Generally lower (e.g., < 0.50), indicating less explained variance	Typically, higher (e.g., > 0.70), indicating a better model fit to data	Schumacker & Lomax (2010); Zhu, Raquel, & Aryadoust (2020)
Chi-square (χ²)	Sensitive to sample size; often shows significant values (e.g., p < 0.05)	Also, sensitive but typically shows lower values with better fit	Zhu, Raquel, & Aryadoust (2020)
Goodness-of-Fit Index (GFI)	Values often < 0.90, indicating poorer fit	Values > 0.90; indicating good fit to the data	Bolton et al. (2018); Zhu, Raquel, & Aryadoust (2020)
Adjusted Goodness-of-Fit Index (AGFI)	Values often < 0.90	Values > 0.90; indicating better fit when adjusting for parsimony	Bolton, et al., (2018); Zhu, Raquel, & Aryadoust (2020)
Comparative Fit Index (CFI)	Often < 0.90, indicating suboptimal fit	Values > 0.90; indicating strong model-data fit	Kruschke (2015); Zhu, Raquel, & Aryadoust (2020)
Root Mean Square Error of Approximation (RMSEA)	Values often > 0.08, indicating poor fit	Values < 0.05; indicating a well-fitting model	Browne & Cudeck, (1992); Zhu, Raquel, & Aryadoust (2020)
Standardised Root Mean Square Residual (SRMR)	Often lacks consistency, higher values (e.g., > 0.10)	Typically, lower (e.g., < 0.08), reflecting the standardised difference between observed and predicted correlations	Kronemann, (2022); Mehedintu & Soava (2022); Zhu, Raquel, & Aryadoust (2020)
Akaike Information Criterion (AIC)	Generally higher AIC values (e.g., > 100), indicating a less optimal model	Lower AIC values (e.g., < 80), reflecting a better-quality model	Zhu, Raquel, & Aryadoust (2020)
Bayesian Information Criterion (BIC)	Higher BIC values (e.g., > 100), less effective in model selection	Lower BIC values (e.g., < 80), indicating a more reliable model	Kruschke (2015); Zhu, Raquel, & Aryadoust (2020)

Review of Dynamic Structural Equation Models for Real-Time Consumer Behaviour: Methodological Advances and Applications Insights

Figures & Tables

Figure 1.

A Comparative framework of traditional and dynamic SEM

Comparative analysis of performance metrics for traditional SEM and dynamic SEM models

Paradigm

My account