Visual working memory (VWM) refers to the active maintenance of visual information needed for higher cognitive processing in the present moment (Luck & Vogel, 2013). Typically, WM is limited to maintaining three to four chunks of information (Cowan, 2001). Like other fluid cognitive abilities, WM declines with age (Craik & Bialystok, 2006; Park et al., 2002). Furthermore, deficits in WM often occur with neurological diseases and psychological disorders. The limited capacity of WM, and its critical involvement in many disorders, has stimulated intensive research efforts into the effectiveness of WM enhancement interventions.
In particular, there is growing interest in affordable and non-invasive brain stimulation techniques, such as transcranial direct current stimulation (tDCS). Benefits of tDCS have been demonstrated for healthy young adults (Fregni et al., 2005; Ohn et al., 2008; Hsu et al., 2014; Johnson et al., 2022; Jones & Berryhill, 2012; Karthikeyan et al., 2021; Tseng et al., 2012; Zaehle et al., 2011), healthy older adults (for a review, see Siegert et al., 2021) as well as clinical cohorts with mild cognitive impairment and early Alzheimer’s disease (for reviews, see Chen et al., 2022; Hsu et al., 2015), attention-deficit hyperactivity disorder (for reviews, see Cosmo et al., 2020; Salehinejad et al., 2020) and major depressive disorder (for reviews, see Brunoni et al., 2012; Woodham et al., 2021).
Typically, tDCS delivers weak currents from anode to cathode through the skull, generating electric fields to modulate cortical activities and facilitate neuroplasticity (Roche et al., 2015; Ruffini et al., 2013). Anodal stimulation is assumed to increase cortical excitability to enhance cognitive functions, whereas cathodal stimulation decreases excitability and, thus, inhibits brain activities (Bikson et al., 2012; Nitsche & Paulus, 2000). However, such polarity-specific effects of tDCS are likely an oversimplification when considering complex cognitive functions like VWM. For example, whereas excitatory effects of anodal stimulation are largely robust, inhibitory effects of cathodal stimulations are less consistent when it comes to studies investigating complex cognition rather than motor effects (for a review, see Jacobson et al., 2012). Taken together, regardless of inconsistent cathodal effects, anodal stimulation has been shown to consistently modulate the neural activities in the target brain regions and, thus, is a promising avenue to enhance the corresponding cognitive functions.
Given that neural activation of frontal-parietal brain regions is known to be involved in the maintenance of VWM representations (Curtis & D’Esposito, 2003; Ikkai & Curtis, 2011), a growing body of research has investigated the possible VWM benefits of anodal stimulation of the dorsolateral prefrontal cortex (DLPFC) and posterior parietal cortex (PPC) (Arciniega et al., 2018; Baumert et al., 2020; Li et al., 2017; Wang et al., 2019). However, some previous studies have showed null effects of tDCS for both DLPFC and PPC stimulation (Dumont et al., 2021; Nikolin et al., 2018; Robison et al., 2017). Some meta-analyses quantifying the effectiveness of tDCS across multiple studies report medium effects of single-session tDCS on VWM (Brunoni & Vanderhasselt, 2014; Dedoncker et al., 2016; Wischnewski et al., 2021), while others report only negligible effects of single-session tDCS (Hill et al., 2016; Horvath et al., 2015; Mancuso et al., 2016; Medina & Cason, 2017). These inconsistencies on the meta-analytic level point to several critical caveats of meta-analyses. Specifically, any conclusions drawn from meta-analysis depend on the included primary studies. First, if the included primary studies largely reported only positive effects, together with overestimated study-level effect sizes (Halsey et al., 2015), it can lead to high false-positive rate of meta-analyses (Kvarven et al., 2020; von Bastian et al., 2019). Second, tDCS studies vary widely in their design, such as administering online or offline protocols (Živanović et al., 2021), stimulating different regions (Wischnewski et al., 2021), or using different VWM paradigms which may require different cognitive processes to one another (Saucedo-Marquez et al., 2013). These methodological variations could have contributed to the inconsistencies observed across both single studies and meta-analyses. Therefore, replications of those studies that reported positive results, using the same parameters and cognitive paradigms, may yield more conclusive evidence as to whether tDCS is effective or not.
The present, pre-registered replication study, therefore, focuses on a particularly high-profile study by Wang et al. (2019) who recently reported selective benefits of anodal tDCS over the PPC, but not DLPFC, on VWM. Wang et al. used a continuous-reproduction VWM paradigm and fitted the mathematical standard mixture-model (Zhang & Luck, 2008) to estimate VWM capacity (quantity of representations maintained in VWM) and precision (quality of those representations). In this task, participants memorised the orientations of 2, 4, or 6 bars on a screen. After either a short (100 ms) or long (1000 ms) interval, participants were asked to reproduce the orientation of one of the bars by mouse-click. The deviation of the reproduced orientation from the original orientation was then used to estimate VWM capacity and precision for each participant, interval duration, set size, and stimulation condition. Wang et al. tested the effects of 15-min anodal tDCS over the left DLPFC and the right PPC relative to a sham condition with a within-subjects design in 20 participants. After excluding two participants due to their poor performance at set size 6, Wang et al. observed a selective increase in VWM capacity for the long retention interval at this set size after PPC stimulation relative to sham, but not after DLPFC stimulation, at any other set size, short retention interval, or on VWM precision.
Wang et al. (2019) interpreted these findings as “causal evidence” (p. 535) of the role of the PPC for VWM functioning. They further argued that “tDCS could be used as promising non-invasive method to enhance [VWM]” (Wang et al., 2019, p. 535). Indeed, Wang et al.’s findings have several significant theoretical and practical implications to the fields of VWM and tDCS. First, the findings from this study falsified the role of the anodal stimulation on DLPFC in improving VWM, thereby contradicting previous studies (Andrews et al., 2011; Fregni et al., 2005; Ohn et al., 2008; Zaehle et al., 2011). Second, by showing that tDCS selectively increases the capacity, but not the precision, of representations held in VWM, Wang et al.’s (2019) findings strongly favour theories conceptualizing the capacity limit of VWM as discrete memory slots (Zhang & Luck, 2008) over those assuming a flexible, continuous resource (Schneegans & Bays, 2016; Van Den Berg et al., 2012). Third, the promising benefits of PPC-tDCS suggests that VWM capacity can be expanded with a non-invasive, cost-effective method, with strong practical implications for clinical tDCS applications. Importantly, by employing a sham-control and through the null effects of DLPFC stimulation, Wang et al. (2019) excluded the possibility that these changes were driven by placebo effects or global excitability with tDCS (Dawood et al., 2019). Furthermore, Wang et al.’s (2019) additional control for sensory memory also ruled out the possibility that these changes were due to mere sensory processes or attentional regulation.
Given these far-reaching implications, it is imperative to ensure that Wang et al.’s (2019) findings are robust and replicable. Replication studies can verify the reliability of the originally reported effects (Simons, 2014), and test the generalizability across conditions that inevitably differ from the original study (Nosek & Errington, 2020). This is particularly critical in the present replication study because, despite its important findings and implications, several aspects of Wang et al.’s (2019) study are potentially problematic and warrant further investigation. First, Wang et al. retained only a small sample of 18 participants for analysis. The small sample size translates into low statistical power even for moderate effect sizes, and low statistical power can lead to false-positive findings (Button et al., 2013). The reported effect size is very large (d = 1.03), but this may reflect an overestimation due to low statistical power (Halsey et al., 2015). Second, Wang et al. (2019) administered only 60 trials per design cell. Such relatively small numbers of trials increase bias and noise variance and thus reduce the precision of the parameter estimation (Lerche et al., 2017; Wiecki & Frank, 2013). Third, the counterbalancing of conditions was likely incomplete in Wang et al. (2019) and, thus, their design did not adequately control for possible carryover (e.g., practice) effects across sessions. Specifically, the study entailed three sessions (DLPFC, PPC, and sham stimulation), resulting in at least six possible sequences requiring counterbalancing. However, with 20 participants completing the experiment (and 18 included in the analysis), it is impossible to assign an equal number of participants to all sequences. Consequently, it cannot be excluded that carryover effects contributed to the previously reported effects. Finally, like many other studies, Wang et al. (2019) used rotated bars as stimuli. Thus, the unique angles of their stimuli effectively ranged only from 0° to 180°, leaving room for developing task-specific strategies. For example, participants may have realised they could simply memorise the location of just one end of the bar rather than the actual orientation of the bar, thereby making the task considerably easier.
To address these potential issues of their study, in this pre-registered experiment, we aimed to replicate Wang et al.’s (2019) study, using a bigger sample size, larger number of trials, complete counterbalancing, and stimuli that use the full space of possible angles (360°). Our pre-registered hypotheses (https://osf.io/n9fkp) based on Wang et al.’s findings were as follows:
Hypothesis 1
PPC stimulation will increase VWM capacity more than DLPFC stimulation. This effect is particularly pronounced at a high difficulty level of the task (i.e., set size 6).
Hypothesis 2
Neither PPC nor DLPFC stimulation will improve VWM precision.