First, descriptive data is presented on the two samples. Then the results of the cluster analysis are described, and the clusters are compared on positive and negative emotionality, life quality and clinical status. After that, the results are analyzed with regard to the research questions: First the hypothesis on distraction is tested. Second, the alternative hypotheses concerning the combination of a high use, and low use, of both acceptance and cognitive restructuring are compared.
Descriptive data on demographic and clinical variables for the clinical and non-clinical samples are presented in Table 1. On the demographic variables, the samples differed significantly with regard to age, t (808) = 4.54, p < .01, and level of education, χ2 (2) = 11.4, p < .01, but not with regard to gender, χ2 (1) = 1.54, p = .21. With regard to positive and negative emotionality and quality of life, the differences between the samples were significant and large for all variables [PANAS-N: t (808) = 15.9, p < .01, d = 1.1; PANAS-P: t (808) = -17.1, p < .01, d = -1.2; Total WHOQOL: t (808) = 19.0, p < .01, d = 1.3]. The bivariate correlations between all self-report measures are presented in Table 2.
As stated above (see the section on hierarchical cluster analysis), the clustering procedure resulted in a nine cluster solution. Table 3 shows the means and standard deviations on each factor for each cluster as well for the whole group, whereas Fig. 1 displays the profile of z-scores (cluster mean minus total group mean, divided by the standard deviation of the total group) for each cluster on the 6 factors.
Comparisons Between the Clusters on Well-Being
A one-way ANOVA with the nine clusters as independent variable and age as dependent variable showed a significant omnibus effect [F(8, 800) = 2.74, p < .01]. Tukey-corrected post-hoc comparisons showed that the only significant differences between the clusters were that Cluster 4 was older than Cluster 3 and Cluster 7. The gender distribution did not differ significantly across the nine clusters [χ
2(8) = 13.9, p = .09]. With regard to level of highest completed education, there was a significant omnibus effect, indicating that there were significant differences in the distribution across the clusters [χ
2(16) = 46.2, p < .01]. To further investigate this effect, the observed frequency in each cell was compared with the frequency expected if the educational level was randomly distributed across the clusters. The statistical testing was performed in accordance with the fixed-margins model using EXACON (Bergman and El-Khouri 1987). After adjusting the alpha level to allow for multiple comparisons using Bonferroni correction (critical α = .002; .05/27) the analysis showed that the only significant effect was that more participants in Cluster 2 had completed a university level education than what was to be expected by chance (observed frequency 54, expected frequency 34.2, χ
2 = 11.5, p = .0008).
Three Bonferroni corrected (critical α’s = .017) ANOVAs were performed with the nine clusters as independent variables and scores on PANAS-N, PANAS-P and WHOQOL as dependent variable in the separate analyses. To control for the observed differences between the clusters with regard to age and level of highest completed education, these variables were entered as covariates in the analyses. Means and standard deviations on the criterion variables for each cluster are presented, in descending order according to the mean score of the clusters, in Tables 3, 4 and 5 (one table for each criterion variable). The omnibus tests showed that there were significant differences between the clusters on all three variables [PANAS-P: F(8, 798) = 85.1, p < .01, partial η2 = .46; PANAS-N: F(8798) = 116.9, p < .01, partial η2 = .54; WHOQOL: F(8, 798) = 110.7, p < .01, partial η2 = .53]. Sidak-corrected post hoc comparisons were performed to examine the significance of the differences between the clusters on each dependent variable. The results from these analyses are presented in Tables 3, 4 and 5.
Representation of the Clusters in the Clinical and non-Clinical Samples
To study the over- and under-representation of each cluster in the clinical and non-clinical samples, the clusters were cross-tabulated with the two samples. Table 7 shows a comparison of the observed frequencies in each cell with the frequencies expected if the clusters had been randomly distributed across the samples. The statistical testing was performed in accordance with the fixed-margins model using EXACON (Bergman and El-Khouri 1987).
The Adaptiveness and Maladaptiveness of Distractive Refocusing
As seen in Table 3 and in Fig. 1, two clusters showed high scores on Distractive Refocusing: cluster 3 and cluster 5. These two clusters had almost exactly the same score on Distractive Refocusing, but showed otherwise very different profiles. Whereas Cluster 3 combined high scores on Distractive Focusing with high scores on Thought Avoidance, Cluster 5 combined high scores on Distractive Refocusing with high scores on Active Acceptance. According to the hypothesis, Cluster 5 should show higher well-being than Cluster 3. This hypothesis was supported on all four variables: Cluster 5 scored significantly higher than Cluster 3 on positive emotionality (see Table 3), significantly lower on negative emotionality (see Table 4), and significantly higher on life quality (see Table 5). In addition, as seen in Table 6, Cluster 5 was under-represented in the clinical sample (observed frequency: 11, expected frequency: 20, χ
2 = 4.0, p < .05), whereas Cluster 3 was over-represented in the clinical sample (observed frequency: 20, expected frequency: 13, χ
2 = 3.6, p < .05).
Combining High Use of Acceptance with High Use of Cognitive Restructuring
As seen in Fig. 1, one of the clusters (Cluster 1) was characterized by consistently high scores on acceptance and consistently low scores on cognitive restructuring, and another (Cluster 2) by consistently high scores on both acceptance and cognitive restructuring. Interestingly, these two clusters showed a very similar profile on the acceptance factors, but differed widely on the cognitive restructuring factors. Still, as seen in Tables 4, 5, 6 and 7, they did not differ significantly on any of the indicators of well-being. Interestingly, although both of these clusters were large (104 individuals in Cluster 1, and 108 individuals in Cluster 2), they were both completely unrepresented in the clinical sample – all 212 individuals were from the non-clinical sample. This speaks against the hypothesis that acceptance and cognitive restructuring have an additive or interactive effect on well-being. The results are, however, consistent with both of the other hypotheses – that is, the equifinality hypothesis (i.e., acceptance and cognitive change strategies achieve similar outcomes), and the “acceptance-is-essential hypothesis”. Unfortunately, the cluster analysis did not identify any cluster with consistently high scores on cognitive restructuring and consistently low scores on acceptance – if so, it would have been possible to contrast these two hypotheses.
Combining Low Use of Acceptance with Low Use of Cognitive Restructuring
As seen in Fig. 1, one of the clusters (Cluster 6) was characterized by consistently low scores specifically on acceptance in combination with about average scores on cognitive restructuring, whereas another cluster (cluster 7) showed consistently low scores on both acceptance and cognitive restructuring. Providing partial support for the “no-strategies-is-worse” hypothesis, Cluster 7 scored significantly lower than Cluster 6 on positive emotionality (see Table 3). However, there were no significant differences between the clusters on negative emotionality (Table 5) or life quality (Table 6), and both clusters showed similar over-representation in the clinical sample and under-representation in the non-clinical sample (see Table 7).