Guidance Required Regarding High Reliability Values in a Higher-Order PLS-SEM Model

[Nikita Singh Dahiya]

Dear Researchers and Experts,

I hope you are doing well.

I am currently working on a PLS-SEM study using SmartPLS involving reflective-reflective higher-order constructs through the repeated indicator approach. I would highly appreciate your guidance regarding a methodological concern related to reliability values in my measurement model.

My higher-order independent construct is “Virtual Influencer Characteristics,” consisting of four first-order dimensions:

• Visual Appeal
• Novelty
• Anthropomorphic Cues
• Technological Sophistication

Initially, after running the repeated indicator approach on a sample of 316 respondents, the higher-order construct produced the following results:

• Cronbach’s Alpha = 0.944
• Composite Reliability = 0.947
• AVE = 0.816

Additionally, one indicator under Visual Appeal showed an abnormally high loading (>1), while one indicator under Novelty had a relatively weaker loading (~0.64). Based on measurement model assessment, I temporarily removed the higher-order constructs and purified the first-order constructs individually.

The following indicators were removed:

• VICVA4 (due to unstable loading above 1)
• VICN1 (due to comparatively weak loading)

After purification and reconstruction of the higher-order construct, the updated results became:

• Cronbach’s Alpha = 0.938
• Composite Reliability = 0.941
• AVE = 0.799

The second-order construct loadings are now:

• Visual Appeal = 0.873
• Novelty = 0.902
• Anthropomorphic Cues = 0.907
• Technological Sophistication = 0.892

Most first-order constructs also demonstrate:

• acceptable AVE values (>0.50),
• acceptable VIF values,
• and satisfactory discriminant validity.

However, I recently received feedback suggesting that reliability values above 0.90 may sometimes indicate redundancy, overly homogeneous data, or possible data quality concerns. This has made me uncertain about whether my current results are still considered acceptable for a reflective-reflective higher-order construct in PLS-SEM.

Therefore, I would sincerely appreciate expert guidance on the following points:

1. Are CR and Alpha values around 0.93–0.94 still considered acceptable in higher-order reflective models using the repeated indicator approach?
2. Would further item deletion be methodologically advisable at this stage?
3. Do the current results appear statistically defensible, or do they still suggest possible redundancy concerns?

I would be extremely grateful for your methodological suggestions and expert opinion.

Thank you very much for your valuable time and guidance.

Kind regards,
Nikita

[Rajinder Kumar]

1. Are CR and Alpha values around 0.93–0.94 still considered acceptable in higher-order reflective models using the repeated indicator approach?

-In many of the research studies alpha is not used, Rho_CR and AVE values are considered robust parameters of construct validity. Also. upto .95 is an acceptable limit (https://pmc.ncbi.nlm.nih.gov/articles/PMC4205511/).

already some discussion https://groups.google.com/g/dataanalysistraining/c/tnU18X59pnU

2. Would further item deletion be methodologically advisable at this stage?

If a researcher sticks to ethics, he/she must not delete/add any items just to fabricate the results, without following the set guidelines. But, if you opt  citations for   Alpha  .95, you don't need to delete any item

3. Do the current results appear statistically defensible, or do they still suggest possible redundancy concerns?

Neeraj sir and Atul Sir can answer it better. 

--
The members of this group are expected to follow the following Protocols:
1. Please search previous posts in the group before posting the question.
2. Don't write the query in someone's post. Always use the option of New topic for the new question. You can do this by writing to dataanaly...@googlegroups.com
3. It’s better to give a proper subject to your post/query. It'll help others while searching.
4. Never write Open-ended queries. This group intends to help research scholars, NOT TO WORK FOR THEM.
5. Never write words like URGENT in your posts. People will help when they are free.
6. Never upload any information about National Seminars/Conferences. Send such information
in personal emails and feel free to share any RESEARCH-related information.
7. No Happy New Year, Happy Diwali, Happy Holi, Happy Birthday, Happy Anniversary, etc. allowed in this group.
8. Asking or sharing Research Papers is NOT ALLOWED.
9. You can share your questionnaire only once.
---
You received this message because you are subscribed to the Google Groups "DataAnalysis" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dataanalysistrai...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/dataanalysistraining/7345f4d5-7c6d-490b-9a44-07e348cd7c5bn%40googlegroups.com.

[Muhammad R Siregar]

Dear Nikita,

Are you sure that your construct is correctly reflective–reflective? With a glance, yours may be formative at the second-order level (higher-order construct, HOC).

If we apply four classic decision rules for measurement specifications (see Jarvis et al., 2003), reflective–reflective assumptions seem to fall apart.

Let us take a look at the direction of causality as the first rule. Does “Virtual Influencer Characteristics” cause an influencer to be visually appealing, novel, anthropomorphic, and technologically sophisticated? No, it does not. The causality goes the other way around.

The second decision rule: Are the dimensions interchangeable? No, they are not. The dimensions are distinct facets that together make a whole (i.e., the HOC).

The third decision rule: Expected covariance in reflective construct. Is a highly “Novel“ virtual influencer expected to be highly “Anthropomorphic”? No, it is not. A highly novel virtual influencer can be a simple geometric shape with very low anthropomorphism.

The fourth decision rule: Nomoogical net. Do these four dimensions share same antecedents and consequences? No, they do not.

Please give it a thought. Good luck with your study.

Best,

MRS

To view this discussion visit https://groups.google.com/d/msgid/dataanalysistraining/CAG9B3sVYeWgV%3DKc8OuXE7cFa4dY%2B7D34Z6EERVQw_magMjXpsw%40mail.gmail.com.