I must say this sounds like a bad approach to me. I can't see how it could make sense to make a selection based on an arbitrary criterion/threshold about regression coefficients in multiple regression. CFA implies redundancy (high correlations) of the measures pertaining to a given factor. When you include multiple highly correlated indicators of the same factor in a multiple regression model, some will have lower regression weights simply because of the (desirable) redundancy (overlap) of the indicators.

As the name implies, CFA is a confirmatory type of analysis. The choice of measures for CFA and SEM should be a result of your theory about the construct(s), not prior empirical selection of variables based on regression or some other technique. So if there is a reference that recommends the above approach, I would be skeptical.

I have come across some papers where the items in a construct were selected based on their standardised regression coefficients for doing the CFA and later the SEM (using SPSS and AMOS).

If the weights were lower they were not selected. Sometimes the items where taken if they were > .6.

In another paper it mentioned taking items if the weight were > .4

(There are many constructs in the research problem, and sometimes each construct has many items. Based on the weights we would like to select some items which have a higher weight.)

I understand that items can be selected based on theory, but when the weight is low it means that in that sample where we are testing it is low because of whatever reason. Thus we select those items which have a higher weight. As I had come across a paper where it mentioned weight > .6 (I missed noting down the reference) so i was wanting to get that reference if at all there is some rule like that....

thank you so much for your response Christian ! answers from experts help a lot. Thanks once again

There is no hard and fast rule for selecting items for confirmatory factor analysis (CFA) or structural equation modeling (SEM) based on standardized regression weights. The threshold for selecting items can vary depending on the specific research question, the construct being measured, and the measurement instrument being used.

That being said, a commonly cited guideline in the literature is to retain items with standardized regression weights of at least .50 or .60 for CFA or SEM. For example, Kline (2015) suggests that standardized regression weights of .50 or higher indicate that an item is a good indicator of a construct, while Hair et al. (2017) suggest a more conservative cutoff of .60 for CFA.

It is important to note that the decision to retain or remove items should not be based solely on standardized regression weights, but also on other factors such as item content, face validity, factor loadings, reliability, and theoretical considerations.

References:

Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2017). A primer on partial least squares structural equation modeling (PLS-SEM). Sage publications.

Kline, R. B. (2015). Principles and practice of structural equation modeling. Guilford publications.

Ma'Mon Abu Hammad Sir,

Thank you so much for your detailed explaination and the reference. It will help me a lot. It is so nice to have experts guiding any one who has a query and hand holding... so that we can reach our destination. Thank you once again sir.....