You may drop as many as indicators from the constructs to improve your construct reliability, but with following conditions:

a) Number of items retained should sufficiently demonstrate / reflect the construct i.e., construct's face validity / content validity should be ensured.

b) the number of items retained should be at least 3 items per construct for CB SEM, however, VB SEM doesn't require any minimum number of items, as it converges with any number of items per construct.

The minimum number of indicators for reflective construct should be four otherwise there will be problem of model identification. 

Thanks for helping me gain understanding. 

As per Hair et al minimum number of indicators in CB SEM can be 3 and for PLS SEM can be 1. Will these both the conditions create problem of  model identification? And also pl tell if we have to retain minimum 4 indicators even in case of PLS SEM?

Though authors have recommended 3 indicators but it is my personal observation that 3 indicator mofel has problem of identification

HI!! Sarita,

Since PLS is based upon principal component analysis, latent construct is also referred to as a component.  Unlike latent construct in CB –SEM, there is no restriction on number of indicators in a reflective model.  In case of CB-SEM< multiple items increase reliability and improve model performance as compared to single-item variable which may cause identification and convergence problems. However, this is not a problem in PLS-SEM, we may have a single-item measure  depending upon the objective.  However, we need to check and mention the indicator/item reliability. Indicator reliability is calculated as the  square of the measurement loading that is .7 *.7 =.49. Hence loading greater than .7 is preferred.

Hope it is clear...

Dear madam,

Thanks a lot. It's really useful.

I have a doubt about the number of indicators that could be dropped as most of the texts approve of dropping not more than 20% of the total number of indicators. Do we follow this criteria in case of PLS SEM (reflective model) too?

Regards

Sarita

Sarita, I told you earlier as well, in PLS we can also have a single indicator, and as such,  you can drop as many indicators that show weak loadings.  You need to give justification that retained indicator reflects the construct well.

HARDIP

Thank you madam. Will follow this for PLS.

Regards

Hi Sarita, can you please share the reference for the PLS SEM , where it says that one indicator would be enough.

Thank you

Please refer to pg 16 of the following book:

A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM) by Hair, Hult, Ringle and Sarstedt (2014).

It clearly mentons that PLS SEM may be applied to constructs measured using single item.

Sarita Sood

With PLS-SEM, we can not delete more than 20% items as per the guidelines provided by Hair et al (2017). Doing so would pose serious questions on content validity of the construct.

Hi, Mr Wali, is there any reference to support this? Waiting for your reply and thank you

Hi, Mr. Wali Ur Rehman. Is there any reference that can support this? Waiting for your reply and thank you...

Wali Ur Rehman, Sir..is that mean we can drop indicator more than 20% as long as they meet these conditions? ;

a) Number of items retained should sufficiently demonstrate / reflect the construct i.e., construct's face validity / content validity should be ensured.

b) the number of items retained should be at least 3 items per construct for CB SEM, however, VB SEM doesn't require any minimum number of items, as it converges with any number of items per construct.

Thanks..

(Hair, 2022). P.57

"Single-Item Measures and Sum Scores

Rather than using multiple items to measure a construct, researchers sometimes choose to use a single item. PLS-SEM proves valuable in this respect as the method does not suffer from identification problems when using fewer than three items in a measurement model as is the case with CB-SEM."