Means and standard deviations for all measures are presented in . Analyses were conceived as the following 4 steps: First, to determine whether the first 4 subscales of the ISMIS (alienation, stereotype endorsement, discrimination experience, and social withdrawal) could be combined into a total score for the purposes of using it in a cluster analyses, internal consistency was determined. This revealed a high degree of internal consistency, coefficient α = .86, which was reduced to .80 if the fifth score was included. Correlations among all 5 ISMIS scores revealed that the first 4 scores were all significantly related to one another (P < .05), while the fifth, stigma resistance, was significantly correlated only with stereotype endorsement. These findings were taken as support to not include the fifth score in the summary score.
Mean and Standard Deviations
In the second phase of analyses, insight and stigma scores were separately correlated with symptoms, social function, hope, and self-esteem. These revealed that internalized stigma was significantly related (P < .05) to positive (r = .36) and negative (r = .26) symptoms, as well as hope (r = −.45), self-esteem (r = .54), and QOLS interpersonal relations (r = −.34). Insight was related to positive (r = .24) and negative (r = .29) symptoms, as well as self-esteem (r = .26).
In the third phase, ISMIS and PANSS insight and judgment items were standardized into z
scores and a K-Means cluster analysis was performed to identify 3 homogenous participant groups based on these scores. Cluster analysis is a method of classifying people into typologies by determining clusters of participants that display small within-cluster variation relative to the between-cluster variation.42–44
In cluster analysis, each participant is assigned to a cluster and participants are moved from one cluster to another until terminating conditions are met. In essence, a cluster analysis is similar in some respects to both factor analysis and discriminant function analysis. It differs primarily from factor analysis in that its end is the determination of orthogonal groups of participants rather than orthogonal groups of variables, and it differs from a discriminant function analysis in that determining group assignment is the goal and not known ahead of time.
K-Means cluster analysis is a nonhierarchical form of cluster analysis appropriate when hypotheses exist regarding the number of clusters contained in a sample. It produces the number of clusters as initially called for, minimizing variability within clusters and maximizing variability between clusters. We chose this procedure rather than rationally defining groups in order to determine, in an exploratory and statistical manner, whether we could detect participants who demonstrated patterns of these scores as hypothesized rather than as we had artificially defined them ahead of time. To give the groups contextual meaning, however, we assigned labels with reference to the meaning of the insight and stigma scores. First, we chose to categorize groups as “low insight” if the PANSS insight item was equal or greater than 4 of 7 and scores of less then 4 as good insight. This categorization has been used elsewhere30
and reflects the difference between general vs minimal awareness that something is wrong. To describe a group's stigma level, we decided that groups with scores of 2 or less would be labeled as “minimal stigma” because such scores indicate general disagreement with items. Scores of greater than 2 but less than 2.5 were chosen to be labeled as “mild stigma,” as these scores reflect agreement of roughly less than half of the ISMIS items. Scores of greater than 2.5 but less than 3 were chosen to be labeled as “moderate stigma” because these scores reflect either agreement of more than half of ISMIS items or strong agreement on several. Scores of greater than 3 were chosen to be labeled as “severe stigma” as these scores reflect either agreement with all ISMIS items or strong agreement on many items.
The cluster analysis produced 3 groups which, based on insight and stigma levels, we have labeled low insight/mild stigma (n = 23), high insight/minimal stigma (n = 25), and high insight/moderate stigma (n = 27). As revealed in , these groups did not differ significantly in age, education, or hospitalization history. Chi-square analyses additionally found that groups did not differ in proportion of participants with schizoaffective disorder or schizophrenia or in proportion of participants from the VA medical center or the CMHC. Analysis of variance also found no significant differences in raw stigma or insight score for participants from the VA medical center compared with the participants from the CMHC. Examination of symptoms, however, found that the high insight/minimal stigma group had significantly lower levels of positive and negative symptoms on the PANSS.
Background and Symptoms Among Groups
In the third phase of analyses we compared hope, self-esteem, and social function between groups. As revealed in , the high insight/minimal stigma group had significantly better interpersonal function on the QOLS than either of the other 2 groups. The high insight/moderate stigma group, by contrast, reported significantly poorer self-esteem and lesser hope than either of the other 2 groups. Given that the groups differed significantly on positive and negative symptoms, analyses comparing social function, hope, and self-esteem were repeated with positive and negative symptom scores included as covariates. In these analyses the groups continued to differ on hope and self-esteem (F2,68 = 4.72, P < .05; F2,68 = 9.35, P < .001), with the high insight/moderate stigma group again in post hoc comparisons having significantly poorer hope and self-esteem. When symptoms were statistically controlled for, however, no significant differences were found between groups on the QOLS interpersonal relations scores (F2,62 = 0.53, P = NS).
Hope, Self-esteem, Symptoms, and Social Function Among Groups