Moderate AVE but low composite reliability

Questions about the implementation and application of the PLS-SEM method, that are not related to the usage of the SmartPLS software.
Post Reply
wilfred
PLS Junior User
Posts: 3
Joined: Wed May 10, 2006 1:48 am
Real name and title:

Moderate AVE but low composite reliability

Post by wilfred »

Hi, everyone, I used an instrument called Gregorc Style Delineator (GSD) in my study. The learning style construct is formed by asking participants to rank 10 groups of words with 4 words in each group from 1 to 4. Two orthogonal dimensions, namely C to A and S to R, are proposed in the theory. As as result, four mediation channels, namely CS, AS, CR and AR, with scores from 10 to 40 are resulted and the sum of the four scores is always 100.

In my model, there is a latent variable learning style (LS) with 4 reflective indicators, CS, AS, CR and AR. As I run the model, the AVE of LS was about 0.5 which is still acceptable BUT the composite reliability was extremely low (0.0008). The loadings of them on LS were CS (0.8385), AS (0.5594), CR (-0.8400) and AR (-0.5185) and cross-loadings were not high. I guess this may be due to the constant-sum problem. I removed the indicator AR and re-run the model. AVE increased to about 0.6 and composite reliability was 0.11. Then I removed one more indicator CR and re-run the model. AVE was about 0.51 and composite reliability was 0.67. Can everyone help to explain this result? I learn from the literature that it is justifiable to remove one indicator due to the ipsative problem. But what about the removal of one more indicator? Is it justifiable, at least in this situation?
viswadatta
PLS Expert User
Posts: 139
Joined: Wed Jul 05, 2006 1:43 pm
Real name and title:
Location: Coimbatore, India

reliability of indicators

Post by viswadatta »

To be accepted, a construct must have both reliability and validity. What you are doing by removing indicators is called purification of constructs (proposed by Churchill in 1979). To be accepted a construct must have alpha of at least 0.7(in some literatures it is said to be 0.6) and ave score of at least 0.5. But please note that you can remove indicators only from reflective constructs and not formative constructs.
Post Reply