A Normative Three-Center Study of 407+404+406 Students
Data Material
Power analyses based on available data suggested a sample size of 400 subjects to cover
90% of the expected empirical variance (1,500 subjects for 95%). Accordingly, our study
was comprised of (1) a "learning sample" of 407 college students from Pasadena (USA);
(2) a second "learning sample" of 404 university students from Lausanne (French speaking
part of Switzerland); and (3) a "test sample" of 406 university students from Zurich
(German speaking part of Switzerland). The three study sites were chosen in such a way that
socio-cultural differences of clinical relevance could be detected. All students were asked
to fill out the 28-item Coping Strategies Inventory "COPE" [Carver et al. 1989; available in
standardized form for 6 languages]
along with the 63-item Zurich Health Questionnaire "ZHQ" which assesses the factors "regular
exercises", "consumption behavior", "impaired physical health", "psychosomatic disturbances",
and "impaired mental health" [Kuny & Stassen 1988; available in
standardized form for 6 languages].
Methods
The intrinsic structure of the COPE instrument was determined by means of Neural Network (NN)
analysis. In particular, we searched for the optimum number of dimensions that were reproducible
across study sites while explaining a maximum of the observed between-subject variance. The
function "crit" with free parameters "N" (number of dimensions/scales) and "Nk" (number of items
that make up the k-th scale; k=1,2, N) served as criterion for the iterative optimization that
simultaneously optimized within- and between-scale association (absolute values):
Upon completion of each optimization step, results derived from the learning sample were verified
through the replication sample so that over-adaptation to the local properties of each single
sample could be avoided. As this algorithm does not distinguish between local and global maxima,
a "random-walk" strategy was applied using 10,000 random permutations as start configurations for
the optimization. All scales were orthogonalized by standard Gauss transformation, normalized (zero
means, standard deviations of 10), and validated by computing the correlation between the resulting
scales on the one hand, and the ZHQ factors "regular exercises", "consumption behavior", "impaired
physical health", "psychosomatic disturbances", and "impaired mental health", on the other. We
estimated empirical variances by systematically evaluating all possible n×(n-1)/2 Euclidean
distances between the "n" subjects’ 28-dimensional feature vectors.