[Rasch] R: Construct Validity, Convergence Validity and Discriminant Validity
Fabio La Porta
fabiolaporta at mail.com
Wed Jun 15 01:21:45 EST 2011
The Rasch modelling framework deals with the internal (or structural)
construct validity of an instrument: the actual responses to the items of
the investigated tool are compared with the responses predicted by a
measurement model (the Rasch model) that tells you how the responses to the
items should be if the tool satisfied the requirement for fundamental
Convergence and discriminant (or divergent) validity are instead external
kinds of construct validity that require the comparison of the total score
of your tool with the total score of other tools, measuring, respectively a
similar or a dissimilar construct. This is usually done by calculating a
simple Spearman correlation coefficient in the case of ordinal data: hence,
these comparisons can not be made within the Rasch framework (personally I
use SPSS but any other statistical package will do).
Before the advent of modern psyschometric methods, the latter methods were
the gold standard in psychometrics. The problem is that if the total score
is biased (as it may well be), these comparisons will be biased too.
This now can be easily overcome by doing the Rasch analysis first and
calibrating your instrument: thus, providing that the adequate fit to the
Rasch model is achieved, the logit estimate of ability derived by the total
score will be free of bias. Hence, you may want to evaluate convergent and
divergent validity by correlating the measure estimates of your
Rasch-calibrated tool with the total score (and/or the measure estimates)
provided by other tools, using the most appropriate correlation coefficient
(Spearman for ordinal data, Pearson for interval-level data).
Da: rasch-bounces at acer.edu.au [mailto:] Per conto di Juanito Talili
Inviato: lunedì 13 giugno 2011 9.20
A: Rasch Forum
Oggetto: [Rasch] Construct Validity, Convergence Validity and Discriminant
Can you share your thoughts regarding how construct validity, convergence
validity and discriminant validity are done using Rasch framework?
Thank you and warm regards.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Rasch