Research Methods in AL

🛑How can I boost KMO value in EFA of questionnaires? I will give you eight strategies here:

Variables= Items

1. Remove Variables with Low Individual KMO Values

Use the Anti-Image
Correlation Matrix to identify variables with individual KMO values below 0.5.

2. Examine the Correlation Matrix

KMO depends on high partial correlations being low and zero-order correlations being high.If many correlations are low (e.g., below 0.3), variables may not be suitable for EFA. So, remove weakly correlated or uncorrelated variables.

3. Combine or Collapse Similar Variables

If two or more items measure the same thing and are highly correlated (r > 0.9), consider combining them or removing redundancies.This can reduce multicollinearity, which negatively affects KMO.

4. Increase Sample Size

A larger sample size improves the stability of correlation estimates, which may help raise KMO. A general rule: at least 5–10 participants per variable, but more is better. (I see that many researchers do not pay attention to this simple rule😊)

5. Conduct Item Analysis or Reliability Tests

Use Cronbach's alpha or Corrected Item-Total Correlation to detect poorly performing items.Removing items with low item-total correlation may improve KMO

6. Use Principal Axis Factoring Instead of Principal Component Analysis (but I always go for PAF😊)

While both are used in EFA, Principal Axis Factoring (PAF) works better with lower communalities and might provide better estimates for factorability.

7. Consider Transforming Variables

If items are highly skewed or non-normally distributed, transformation (e.g., log or square root) might reduce noise and improve correlations, potentially improving KMO.

8. Theoretically Reconsider Item Set

Review whether all items conceptually belong together and reflect potentially common underlying factors.

Remove or revise items that are theoretically inconsistent.

🛑I say that from the word go, you need to determine whether your factors are formative or reflective, as many researchers easily overlook this super important distinction 😊

www.tgoop.com/thesisdissertationws/977

792 viewsHessameddin Ghanbar, edited  May 15 at 12:30

🛑How can I boost KMO value in EFA of questionnaires? I will give you eight strategies here:

Variables= Items

1. Remove Variables with Low Individual KMO Values

Use the Anti-Image
Correlation Matrix to identify variables with individual KMO values below 0.5.

2. Examine the Correlation Matrix

KMO depends on high partial correlations being low and zero-order correlations being high.If many correlations are low (e.g., below 0.3), variables may not be suitable for EFA. So, remove weakly correlated or uncorrelated variables.

3. Combine or Collapse Similar Variables

If two or more items measure the same thing and are highly correlated (r > 0.9), consider combining them or removing redundancies.This can reduce multicollinearity, which negatively affects KMO.

4. Increase Sample Size

A larger sample size improves the stability of correlation estimates, which may help raise KMO. A general rule: at least 5–10 participants per variable, but more is better. (I see that many researchers do not pay attention to this simple rule😊)

5. Conduct Item Analysis or Reliability Tests

Use Cronbach's alpha or Corrected Item-Total Correlation to detect poorly performing items.Removing items with low item-total correlation may improve KMO

6. Use Principal Axis Factoring Instead of Principal Component Analysis (but I always go for PAF😊)

While both are used in EFA, Principal Axis Factoring (PAF) works better with lower communalities and might provide better estimates for factorability.

7. Consider Transforming Variables

If items are highly skewed or non-normally distributed, transformation (e.g., log or square root) might reduce noise and improve correlations, potentially improving KMO.

8. Theoretically Reconsider Item Set

Review whether all items conceptually belong together and reflect potentially common underlying factors.

Remove or revise items that are theoretically inconsistent.

🛑I say that from the word go, you need to determine whether your factors are formative or reflective, as many researchers easily overlook this super important distinction 😊

BY Research Methods in AL