Pengenalan Generalised Structured Component Analysis (GSCA)

Summary

The video introduces Generalized Structured Component Analysis (GSCA), explaining its evolution from previous methods like CBSM and PLS. GSCA offers significant advantages, such as not requiring normality assumptions and handling small sample sizes effectively. It contrasts with traditional SEM models, highlighting key differences in factor structures, loading factors, and regression coefficients. The video also details the process of running GSCA in software, interpreting output, and comparing results with other methods like CB-SEM. Overall, GSCA provides a versatile tool for researchers with small datasets or non-normal data distributions.

Takeaways

• 😀 GSCA is a rapid development of earlier SEM methods like CBSM and PLS, and it combines the strengths of CFA with regression models.
• 😀 SEM typically involves three streams: covariance-based SEM (CBSM), partial least squares (PLS), and Generalized Structured Component Analysis (GSCA).
• 😀 GSCA uses a composite model where the arrows go from item to factor, as opposed to CFA, where arrows go from factor to item.
• 😀 GSCA does not require the assumption of normality for each item, making it more flexible for data with violations of normality.
• 😀 One advantage of GSCA is that it avoids common issues like non-convergence, improper solutions, or matrix definiteness problems.
• 😀 The estimation method used in GSCA is Alternating Least Squares (ALS), which is specifically designed for this approach.
• 😀 GSCA allows for factor scores to be generated for further analysis, such as regression, without requiring assumptions of normality.
• 😀 Unlike traditional SEM methods, GSCA can handle smaller sample sizes (e.g., 100 participants) without causing issues like model non-convergence.
• 😀 GSCA's results can be compared to total score regression, showing differences in regression coefficients and error estimates.
• 😀 In GSCA, residual correlations are expected between items, which is different from CFA where these correlations are ideally zero.
• 😀 GSCA is advantageous for studies with small samples, and it can produce valid results even when normality assumptions are violated.

Q & A

What are the three streams in SEM (Structural Equation Modeling)?

The three streams in SEM are Covariance-based SEM (CBSM), Component-based SEM (PLS), and Generalized Structured Component Analysis (GSCA).

What is the main difference between CFA and GSCA in terms of factor to item relationships?

In CFA (Confirmatory Factor Analysis), the relationship is from factor to item (reflective measurement), whereas in GSCA, the relationship is reversed, from item to factor (composite model).

What is the key characteristic of the GSCA estimation method?

The GSCA estimation method uses Alternating Least Squares (ALS), which is specifically designed for GSCA models and does not require normality assumptions for the data.

What is the role of residual correlations in GSCA compared to CFA?

In CFA, residual correlations should be absent, with the goal of maximizing model fit. In GSCA, residual correlations are calculated and included in the model to account for variance that is not explained by the latent variables.

Why is GSCA preferred when sample sizes are small?

GSCA is less prone to convergence issues with small sample sizes and can produce reliable estimation results even with as few as 100 samples, unlike other methods like PLS or CBSM, which may fail to converge.

What is the advantage of GSCA over other SEM methods like PLS or CBSM regarding normality assumptions?

GSCA does not require the assumption of normality for each item, making it more flexible when dealing with non-normal data compared to PLS or CBSM, which rely on normality assumptions for accurate estimations.

How does GSCA handle factor scores, and how can they be used in further analysis?

GSCA generates factor scores, which are the weighted sums of observed variables that can be used in further analysis, such as regression, without needing to assume normality in the data.

What are some common fit indices reported in GSCA?

Common fit indices in GSCA include SRMR (Standardized Root Mean Square Residual), GFI (Goodness of Fit Index), and R-Square, which measure the proportion of variance explained by the model.

What is the significance of the regression coefficient in GSCA, and how does it relate to the R-squared value?

The regression coefficient in GSCA represents the strength of the relationship between variables, while the R-squared value indicates the proportion of variance explained by the model. Both are important for assessing model fit and validity.

When should GSCA be used instead of traditional SEM methods like CBSM?

GSCA is ideal for small sample sizes, when normality assumptions are violated, or when the focus is on a composite measurement model. It is particularly useful when data are not normally distributed or when simpler models are needed for comparison with other methods.