3 A meta-analysis on the association between EA and BPP 85 2.5 Meta-analysis Comprehensive Meta-Analysis (CMA; Biostat, Englewood, NJ) software was used to perform the statistical analysis. Some studies reported means and standard deviations or related statistics (t-test). Other studies reported correlation coefficients (Pearson correlation or regression coefficients). Most studies did not control for possible confounding variables. For studies that did control for confounders, the best available statistics controlling for these confounders were used. CMA converted all measures to correlation coefficients. Cohen’s convention for interpreting the magnitude of correlations (Cohen, 1988) was used: .10 < r < .30 corresponded to a small correlation, .30 < r < .50 moderate, and r > .50 strong. As the types of design, populations, and measures differed across studies, a random effects model was used. Compared to a fixed effects model, this model does not assume that each study is an exact replication of the other studies. Rather, it assumes that the variance in outcomes consists of both true variance and random error. The outcomes of the random effects model are conservative as confidence intervals are generally broad. The analysis was carried out in four steps. First, the overall correlation across all studies was estimated. Second, heterogeneity was assessed, using the Q-statistic and the I2 index (where 0–30% is low, 30–75% is moderate, and 75–100% is high [68]. In addition, we checked whether leaving out the studies with the most extreme correlation resulted in substantial reduction of heterogeneity. Third, sensitivity analyses were performed with respect to four items: design, measurement instruments, specific deficits in alexithymia, and possible negative affect. For design, we analyzed BPD versus healthy controls (HC), BPD versus psychiatric controls (PC), the correlation of borderline personality traits with emotional awareness in a healthy sample, and correlation of borderline personality traits in a patient sample. The analysis for measurement instruments included emotional awareness focus, for example, subscales of the SEIS, TMMS, and Multidimensional Experiential Avoidance Questionnaire (MEAQ) versus alexithymia focus, for example, TAS-20 or BVAQ. Analysis of specific deficits in alexithymia included difficulty in identifying emotions, difficulty in describing feelings, and externally oriented thinking. As a final step, an analysis of a possible spurious effect due to negative affect (NA) was conducted since measures of NA have been found to correlate substantially to measures of alexithymia. For the last step in our overall analysis, we examined publication bias. Three indices were used to assess whether positive findings were over-reported [69]. The first index was a result of our plotting the correlation against its standard error in a funnel plot. When the plot is symmetrically distributed, there is no publication bias. The
RkJQdWJsaXNoZXIy MjY0ODMw