Page 1 of 1

Max: MVs can be removed

Posted: Thu Feb 10, 2011 1:13 pm
by maqsoodmemon
Hi Everybody.
This is my fist time to post question on this forum, would highly appreciate if any body
Could suggest and make comments, whether my approach of assessment of measurement and structural model is right by deleting the low lading MVs.

My model has 4 constructs i.e A (exogenous), B (Mediating),C (Mediating),D (Endogenous)
Construct A measured by 19 MVs
Construct B measured by 14 MVS
Construct C measured by 15 MVs
Construct D measured by 4 MVs

My study is exploratory and majority of items are new developed. Indicators reliability (loading >0.5), cronbach Alpha(all LV 0.9) and composite reliability (all LV 0.9) was ok where as validity was not ok AVE (less than 0.5 for 2 LV) and sq root of AVE in diagonal was not as required as well. It was found some items loading were higher to other constructs than own construct and were deleted. Some more items were removed which has marginally higher (less than 0.1 deference between loadings) loading compare to other LV were also deleted and finally measurement model's convergent and discriminant validity became acceptable values as per criteria (Chin 1998).
Finally Measurement Model passes all tests with following no. of MVs left
A 10 MVs, B 11, C 9 and D 4

structural model shows significant relationship 3 out of 5. After deleting 2 more MVs of construct B and C (those has loadings <0.6) structural model shows significant value for 4 relationship.
Maqsood-University of South Australia

Posted: Fri Mar 30, 2012 1:20 pm
by Diogenes

One year after…

As you said, this is an exploratory procedure, and it is ok.

In a second step (confirmatory analysis) a new sample should be used.

Best regards,