Introduction

• Forms of validity
• Comparing constructs
• Examples from marital assessment
I. Forms of validity
Content:  Adequately sample the content domain (e.g., love)_
Construct:  Correlated with similar construct (e.g., 2 measures of
                 satisfaction)
Criterion:  Outside evidence of construct (e.g., either behavior or another
                test; some clear example of the construct)
Concurrent:  Measures taken at the same time
Predictive:  Measures time 1 predict to construct time 2
                    See here for a good example of  predictive validity.


II. Comparing constructs

Methods:  How traits assessed (e.g., self-report, MICS, etc.)
Targets:  Focus of data
Traits:  Presumed stable, marital behaviors

        Table 2: Hetero-trait hetero-method matrix

                                     ________________________________________________________

    Notice first how each method (1, 2, and 3) is cross-tabulated (nested)
with the other methods under each trait. The r =  .66 (correlation, upper left
corner) is between Method 1and Method 2 for the same trait (i.e., Conformity).
   This pattern --whereby the scores derived by each method are correlated with
the scores of other methods, for each trait-- is true for all the traits listed.
(Note: Although there would also be a correlation coefficient, r, for each cell
in the table, for a total of 27 different rs,we are only showing rs that are
significant.) This approach allows us to see whether by using different methods
we obtain the same "reading" on the trait in question.

High r's within each triangle (in the table) mean that the different methods all
intercorrelate when measuring  the same trait. For example, when measuring
CONFORMITY, by three different methods, a high correlation means that the
methods converge. CONFORMITY --whatever it is-- can be measured
independentlyof the methods! That means we have a construct that is NOT
method dependent, which is a good thing.

    If  however the methods correlated with one another across different
traits (i.e., regardless of the trait being measured), we would have to conclude
that the methods were more similar to each other regardless of what they
were measuring.That is not a good thing! Thus for example,  if method 1 gives
the same result regardless of what it is measuring, (e.g., the same magnitude of
r regardless of which trait it was used to measure) it would mean that what we are
measuring is the performance of the method, not the behavior or trait of interest.
Not a good thing!

 

Return to Main