Discriminant validity

[Diogenes]

Some times, the Fornell/Larcker' method didn´t show discriminant validity. Another way to assess it will be:
Compute the Interval of confidence of the correlation (r +/- 2 * se), if the value 1 is not inside of this I.C. we have discriminant validity.

For this computation we need the s.e. of the LV correlations, it is a good idea because we knew if this correltaions were significant too.

In bootstrap procedure this s.e of the LV correlations could be computed.
Best regards
Bido

[volkerseiler]

Dear Professor Bido,

I wonder whether there is any literature concerning the alternative way of assessing discriminant validity.

Kind regards,
Volker

[Diogenes]

Hi Volker,

1) Cross loadings - the indicator should has a bigger correlation with its LV than with others LV.

2) NETEMEYER, R.G.; BEARDEN, W.O.; SHARMA, S. (2003). Thousand Oaks: Sage Publications.
p.154 = "If the confidence interval ( + 2 standard errors) around the disattenuated correlation does not contain a value of 1, evidence of discriminant validity exists (Anderson & Gerbing, 1988).

The desattenuated correlation = see
NUNNALY, J. C.; BERNSTEIN, I. H. Psychometric Theory. 3rd Ed. New York: McGraw-Hill, Inc., 1994.

Best regards,

Bido

[volkerseiler]

Dear Professor Bido,

Thank you very much.

Kind regards,
Volker

[Plínio Rafael Reis Montei]

We should always remember that correlations distributions are not symmetrical when r differs from 0 (H0 in correlation tests). We should calculate the non-symmetrical confidence intervals using Fisher Z approach. Excel could do it. You need just to inform the original r and then calculate the confidence interval as a usual Z score and then turn back to the original (non symmetrical) interval.

Cheers