|
Correlations |
|
A correlation is used to
provide an indication of the
similarity between two constructs. For
rated grids, the Pearson product-moment correlation is used, for ranked
grids the Spearman rho
correlation is used. In general these coefficients are fairly similar
irrespective of the nature of the grid data. They have an advantage in
that they are independent of the scale of rating, mean rating or
standard
deviation of ratings and form values that lie from –1.0 (where the
constructs
are perfectly reversed in their pattern of ratings) through zero, where
there is no similarity, to +1.0 where they agree perfectly, although
correlations
for dichotomous ratings may not have a maximum value of 1.0 (see
Carroll,
1961). They differ from distances,
in that they only attend to the profile similarity across elements, and
do not take account of differences in level (means) or scatter
(variances).
An old (but readable) discussion of this can be found in Cronbach and
Gleser
(1953). Correlations form the basis of some measures of cognitive
complexity, such as intensity.
They should not be routinely calculated for elements since they will
change
if construct poles are reversed, although this can be overcome in some
computer
programmes such as GRIDSTAT. |
|
References
|
|
- Carroll,
J.B. (1961) The nature of data, or how to choose a correlation
coefficient. Psychometrika, 26, 347-372.
- Cronbach,
L.J., and Gleser, G. (1953) Assessing similarity between profiles. Psychological
Bulletin, 50, 456-473.
|
|
Richard C. Bell
|
|
|