determining discriminant validity

[fried]

i want to determine the discriminant validity in my model. this to ensure that the measures of contructs differ from each other.

theory points me to the fact that AVE must be greater than the variance shared between the constructs and other contructs in the model (squared contruct correlation?).

what values from the report should i use?
* PLS - quality criteria - AVE
and
* PLS - quality criteria - Latent variable correlations?

[Diogenes]

Hi Michiel,
you must use both of them

It's common report this analysis like this:

LV1 0,88
LV2 0,35 0,90
LV3 0,28 0,45 0,75

This is the matrix of correlations between LVs, with the root squared of AVE in the diagonal.
In this case, we have RSqr of AVE > Correl, them there is discriminant validity.
Best regards.
Bido

[fried]

is it possible all the square roots are 1.0000000?

Code: Select all

  
Latent Variable Correlations

COMMITMENT DISTRJUST GENERAL SATISFACTION GENERAL SATISFACTION * Klacht 
COMMITMENT 1,000000       
DISTRJUST 0,309769 1,000000     
GENERAL SATISFACTION 0,618235 0,471873 1,000000   
GENERAL SATISFACTION * Klacht -0,122507 -0,063180 -0,116580 1,000000 
GENERAL SATISFACTION * Klacht -0,118761 -0,058339 -0,115453 0,997454 

[Diogenes]

Hi Michiel,
your output has just correlations (diagonal = 1s).

SmartPLS don't have a matrix done like I had showed above.

You should copy it to Excel, compute the RSq of AVE, and copy these results to matrix of correlations, where there were 1s.

Best regards.
Bido

[fried]

in other words:

replace all the 1.000000's with the corresponding item's AVE^2

[Diogenes]

in fact

Root Squared of AVE (AVE^0.5).

[viswadatta]

Find the RMS values for the ave scores for two constructs. If this value is greater than the construct correlations, the constructs have discriminant validity.