We appreciate Berryhill and Jones' suggestion of including an independent measure for visual working memory capacity, which should serve as the basis for dividing the participants into a low- and high-capacity group. Although the most optimal alternative to the change detection paradigm remains to be identified, we do agree that the use of a relevant and independent measure is a helpful approach for future tDCS studies....
We appreciate Berryhill and Jones' suggestion of including an independent measure for visual working memory capacity, which should serve as the basis for dividing the participants into a low- and high-capacity group. Although the most optimal alternative to the change detection paradigm remains to be identified, we do agree that the use of a relevant and independent measure is a helpful approach for future tDCS studies. Regarding the possibility of regression to the mean in our analysis in Exp 2, we have several observations that we think should alleviate such concern. We list these observations and our rationale below in three points:
1) The improvement effect of tDCS was not present in the sham or M1- tDCS control condition from Exp 1. This was also true when we applied the same median split to our participants based on their sham performance (low K = 2.3, high K = 4.4) and reanalyzed the M1-tDCS data. Thus, the tDCS effect we observed was specific to rPPC, but not M1, making it less likely to be driven by regression towards the mean.
2) To test whether our behavioral results from Exp 2 was simply driven by the way we categorized our participants, we also reanalyzed our rPPC-tDCS data from Exp 1 using the same median split from the sham condition. We observed significant improvement in low-performers (p=.028), as well as nonsignificant improvement in high-performers (p=.1). These low -performers shared the exact same capacity estimates (K=2.3) as the low- performers from Exp 2, thus an rPPC-specific (but not M1 using the same method) improvement makes sense. More importantly, a regression-like effect that is caused by median split would predict a decrease (even if not statistically significant) in high-performers in the rPPC-tDCS condition, but we observed the opposite. Thus, we think the conflicting trends (both nonsignificant) in high-performers across Exp 1 and 2, using the same setup/stimulation, suggests that the directionality of these nonsignificant fluctuations are best treated as nonsignificnat noise instead of evidence for regression to the mean.
3) Lastly, our ERP results from Exp 2 also did not show any trends of regression towards the mean. That is, the amplitude of N2pc and SPCN components increased for low-performers, but did not decrease for high- performers (Fig 4 of Tseng et al., 2012), consistent with the idea from Pt.2 above.
Based on the points above, we stand by the results in our original report, although we agree that they could have been more complete with the aid of a suitable independent measure of visual working memory capacity. To this end, we thank Berryhill and Jones for their helpful suggestion that will benefit future studies investigating the interaction between brain stimulation and people's natural ability and also provide the benefit of more readily enabling comparison of the performance levels of groups across studies.
None declared
Tseng and colleagues should be commended for combining tDCS and ERP approaches in their recent paper. We paid particular attention to the paper because we conducted similar studies with different results (Jones et al. 2012). Where they found that anodal tDCS to right parietal cortex improved working memory performance in low performing individuals, we found a benefit in high capacity participants.
One possible...
Tseng and colleagues should be commended for combining tDCS and ERP approaches in their recent paper. We paid particular attention to the paper because we conducted similar studies with different results (Jones et al. 2012). Where they found that anodal tDCS to right parietal cortex improved working memory performance in low performing individuals, we found a benefit in high capacity participants.
One possible explanation for this difference is that their conclusion may be driven by an analysis that is vulnerable to regression to the mean. Specifically, they performed a median split based on working memory performance during sham stimulation. Thus, the performance difference they observed during active stimulation may be driven by twice measuring working memory in the same task and using one of these measures to form groups. Consistent with a regression to the mean explanation, Tseng et al. report a statistically significant improvement in their low performance group and a numerical but non-significant decrease in their high performance group during the second measure of performance. In contrast, we defined high and low capacity groups using an independent measure of working memory and found that tDCS benefited high capacity participants. Compellingly, re-analysis of our data using Tseng's approach replicates their pattern of results with much of the effect (64%) attributable to regression to the mean. To allay concerns, Tseng et al. might consider collecting an independent measure of working memory capacity and reanalyzing their data or calculating the contribution of regression to the mean and reassessing the direction of their effect.
Sincerely,
Marian Berryhill and Kevin Jones University of Nevada Reno, NV, USA 89557 mberryhill@unr.edu
Reference:
Jones KT, Berryhill ME (2012) Parietal contributions to visual working memory depend on task difficulty. Frontiers in psychiatry / Frontiers Research Foundation 3: 81.
None declared