Differential response latencies

Because all other considered studies used stimulus onset asynchronies greater than 2400ms, only the long SD condition in Matthews and Southall's (1991) study was considered in the aggregation of effect sizes. Weighted mean effect sizes suggest that in general, nondepressed people were relatively facilitated in their responses to positive words with respect to negative or neutral words. These differences appear to diminish when differences are standardized to reflect within-sample variability. Specifically, the weighted mean difference in reaction times to negative and neutral words, D.= -2.06ms, with an average conditional variance, V.= 17.71. The 95$\%$  confidence interval around this mean ranges from D._lb=-10.31 to D._ub=6.18. Because the confidence interval includes zero, it may be stated that D. does not statistically significantly differ from zero. Similarly, the standardized effect size d.= -.022, with the 95% confidence interval extending from d._lb=-.29 to d._ub=.24. In contrast, the weighted mean difference in reaction times to positive and neutral words, D= -16.35ms, V.= 66.13. The $95\%$ confidence interval around this mean ranges from D._lb=-32.30 to D._ub=-0.42. Because this interval does not include zero it may be said that the mean difference of positive to neutral words differs statistically significantly from zero. Yet, the standardized effect size is rather small, d.=-.24, and the 95$\%$ confidence interval does include zero, d._lb=-.54, d._ub=.06. Similarly, nondepressed people's reaction times to positive words were statistically significantly faster than their reaction times to negative words, D.= -20.96ms, V.=54.94, D._lb=-35.48, D._ub=-6.43, and the standardized effect size does reflect this difference, d.=-.26, d._lb=-.52, d._ub=-.01.

Given the large variability in variance estimates between studies as shown in the large range of difference scores and associated variances in Appendix A, it is not unexpected that results using a standardized measure of effect size would not reflect differences apparent under an unstandardized measure of effect size. A homogeneity test reveals that the obtained weighted standardized effect sizes did not differ statistically significantly more from each other than would be expected by chance for the negative v. neutral condition, Q=5.11, df=4, p>.05. For the positive v. neutral condition, Q=1.77, df=3, p>.05 For the positive v. negative condition, Q=0.44, df=5, p>0.05. Caution should be used in interpreting the nonsignificance of these findings, as the Q statistic may not approximate a chi-square distribution for very small n.

Additionally, there is a great deal of variability in the mean reaction times obtained in the studies. This finding suggests that differences in methodologies such as variations in Stimulus Duration (SD), selected words, and criteria for lack of depression may have played a role in the variability between studies. A theory which accounts for differential performance with respect to different valences may therefore be well advised to take these factors into account.