Electrophysiological studies on frontal alpha asymmetry (FAA) in depressive disorders (DD) have almost 40 years of history, with first reports presented in 1983 (Schaffer et al., 1983). Since then, many studies have reported relatively higher alpha band power in the left vs right frontal channels (left-sided FAA) in subjects suffering from DD compared to healthy individuals (Allen et al., 2004; Davidson, 1984; Davidson, 2004; Kemp et al., 2010; Schaffer et al., 1983). FAA index, calculated by subtracting the left-side alpha power from the respective right-side channel, is one of the most common electrophysiological indicators of DD in the current literature (de Aguiar Neto and Rosa, 2019). However, multiple studies failed to replicate the relationship between FAA and DD (Allen et al., 2004; Carvalho et al., 2011; Deldin and Chiu, 2005; Gold et al., 2013; Kaiser et al., 2018a; Kentgen et al., 2000; Knott et al., 2001; Mathersul et al., 2008; Szumska et al., 2021; Vuga et al., 2006) and conclusions of meta-analyses remain skeptical (Thibodeau et al., 2006; van der Vinne et al., 2017). In this light statements about FAA being a biomarker of depression (Baskaran et al., 2012; Iosifescu et al., 2009) seem to be too far-fetched.

It is not clear what the causes of above-mentioned inconsistency in the literature are, but methodological issues are mentioned as one potential problem in a recent review by Kaiser et al., 2018b. Factors like age, gender, or education are known to covary with depression (McFarland and Wagner, 2015; Nolen-Hoeksema, 2001; Stordal et al., 2003) and alpha power or asymmetry (Jesulola et al., 2017; Parameshwaran and Thiagarajan, 2019; van der Vinne et al., 2017), but are often not controlled for during recruitment. While counter-balancing groups with respect to these variables can be difficult their effect may be accounted for by including them as predictors in the regression model. Although correlated predictors like education and depression can reduce each other's effect by explaining shared variance in the dependent variable, including confounding variables can also help remove variance unexplained by the variable of interest and therefore increase its effect.

There are also other important methodological problems worth considering when trying to resolve the validity of FAA in DD. Although much attention in the FAA literature has been paid to the choice of EEG reference (see for example: Smith et al., 2017; or Stewart et al., 2014) other aspects of signal processing and analysis seem to be more neglected. Many EEG studies on FAA use and report FAA index calculated only for a few channel pairs (e.g. one or two pairs were used in 12 out of 17 studies [70.6%] included in the meta-analysis by van der Vinne et al., 2017). In combination with the fact that topographical maps of effects are rarely presented (4/17 studies [23.5%] in van der Vinne et al., 2017) this significantly reduces the reliability and interpretability of the reported effects. The FAA effects are frequently assumed to reflect frontal sources of alpha oscillations but without a topographical map to support this claim it is difficult to conclude whether such interpretation is correct. For example, alpha asymmetry at frontal channels may, in principle, arise due to asymmetrical projection from other, non-frontal, sources. This could be identified in the topography, but not at the single channel pair’s level. Without a topographical map it is also more difficult to assess the physiological reliability of the reported effect – significant effect on one channel pair without similar effects on surrounding channels calls for skeptical consideration (van Ede and Maris, 2016). Therefore, it might be better to perform the analysis on many frontal channel-pairs with relevant correction for multiple comparisons (for example, with the very popular cluster-based permutation approach, Maris and Oostenveld, 2007). This is unfortunately rarely done in FAA studies on DDs (0/17 studies in van der Vinne et al., 2017).

However, performing the analysis only at the channel level, when the research question pertains to the neural source of the effects, can also lead to misinterpretations. Even if topographies are shown, they can be inconclusive with respect to the underlying neural source. For this reason it might be useful to perform source localization and continue the analyses in the source space. Given the assumption of frontal alpha sources of FAA presented in the literature, source level analysis would be appropriate. Regrettably, most FAA studies do not perform source localization, although there are notable exceptions (for example, Lubar et al., 2003; Smith et al., 2018).

Incompatible results in FAA literature, summarized briefly above, suggest that the FAA relationship with DD is sensitive to the choice of signal preprocessing and analysis steps. In such a case applying multiverse analysis (Steegen et al., 2016), that is, presenting results of multiple justifiable analysis paths, is a valuable tool to test the robustness of the studied effects. Multiverse analysis seems to be especially well suited for neuroscience research, given the multitude of preprocessing and data analysis choices that result in a complex ‘garden of forking paths’ (Gelman and Loken, 2014). As most neuroscience studies test only one analysis variant it is difficult to assess the robustness of any individual effect, and it seems that at least some neuroscientific findings are sensitive to the choice of signal analysis steps (Cohen, 2015; Cohen and Gulbinaite, 2014; see also Botvinik-Nezer et al., 2019).

The purpose of this article is to thoroughly test the robustness and credibility of FAA as a marker of DDs and address the limitations of FAA research methodology by performing a multiverse analysis of data coming from five independent studies. We performed 270 analyses in total differing in: (a) the signal space used (channel space vs source space); (b) subselection of the signal space (channel pairs vs all frontal pairs with cluster-based correction); (c) statistical contrast used (group contrasts vs linear regression); (d) statistical control for confounding variables (gender, age and education). Finally we perform analyses on data aggregated across studies and propose additional guidelines to improve quality and reliability of the data analysis in FAA research.

To investigate the validity and robustness of FAA as a marker of DDs we used the multiverse approach (Steegen et al., 2016) and performed a total of 270 analyses of eyes-closed resting EEG recordings (total N = 388) from five independent studies. These data sets differ in EEG recording equipment, cap layout, and characteristics of subject groups (see Figure 1 and Materials and methods, sections: Participants and Electrophysiological data sets).

Figure 1

Download asset Open asset

The analysis variants making up the multiverse analysis can be classified along four major dimensions (Figure 2): (a) statistical contrast used: group comparisons or testing for a linear relationship, (b) the signal space used: channel space (average reference – AVG: 120 analyses, 44%; current source density reference – CSD: 120, 44%) or source space (DICS beamforming, 30, 11%); (c) subselection of the signal space: channel pairs (120, 44%), all frontal pairs with cluster correction (60, 22%) or all frontal channels with cluster-based correction and standardization instead of subtraction (60, 22%; see Signal analysis section); and (d) statistical control for confounding variables (135 without and 135 with control for confounds).

Figure 2

Download asset Open asset

We used four different statistical contrasts in the analyses: two group contrasts using independent t-tests to compare FAA between groups; and two linear contrasts using linear regression to test the relationship between psychometric depression score and FAA. Group contrasts included: comparison between diagnosed and healthy controls (DvsHC) or sub-clinical and healthy controls (SvsHC). The inclusion of SvcHC contrast is motivated by the fact that in some FAA studies depression is not diagnosed by conducting a structured clinical interview – instead groups are created based on score thresholds from psychometric depression questionnaires (for example, De Raedt et al., 2008; Imperatori et al., 2019; Schaffer et al., 1983). For group contrasts we used Welch t-test, which does not assume equal variance of the compared groups (Delacre et al., 2017). Linear contrasts were performed either for all subjects together (allReg) or only for the diagnosed subjects (DReg). allReg contrast quantifies the linear relationship between FAA and BDI across all participants while DReg contrast tests whether FAA increases with depression severity measured with BDI questionnaire (or PHQ-9 in Study V).

Combining all the analytical pathways (studies × statistical contrasts × analysis spaces × analysis approaches × control for confounds) leads to 270 analyses, the results of these analyses are summarized in Tables 1–8.

Channel-pair analyses

We report results of all the channel-pair analyses in Tables 1 and 2. Only 3 out of 60 analyses without control for confounds gave significant results. This is expected by chance, p=0.583, binomial test for a probability of significant result greater than 5%. For analyses controlling for confounds only 1 out of 60 was significant, which is also expected by chance, p=0.954, binomial test. We focus the description below on results from the analyses without control for confounds.

Two of the significant results were found for the diagnosed vs healthy controls contrast (DvsHC) for the average referenced (AVG) F3–F4 channel pair in Study I and V. In Study I right–left alpha asymmetry was lower for the diagnosed group than for healthy controls, t = −2.073, p=0.043 (Figure 3A). However, Study V showed the reverse: alpha asymmetry was higher for the diagnosed group compared to healthy controls, t = 2.823, p=0.007 (Figure 3B). Because we use right minus left alpha asymmetry only results with more negative asymmetry values in depressed individuals (negative t values) are congruent with the classical FAA effect.

Figure 3

Download asset Open asset

We now examine the FAA-congruent result from Study I in a wider context, comparing it to the outcome of related analyses that differ in a single parameter. Applying CSD instead of average reference to the same channel pair did not reveal a significant result (same study, same contrast, F3–F4 pair, CSD reference, t = 0.132, p=0.896). The results were also not significant for the other channel pair, irrespective of reference (same study, same contrast, F7–F8 pair: average reference, t = −0.365, p=0.717; CSD reference, t = 0.553, p=0.583). Performing the same analysis on data from Study III or IV also did not give rise to a significant outcome (same contrast, F3–F4 pair, AVG reference, Study III: t = 0.904, p=0.371; Study IV: t = 0.450, p=0.654). For Study II the DvsHC contrast was not available, but conceptually closest contrast – Sub-clinical vs Healthy Controls (SvsHC) – was found insignificant for both channel pairs using either AVG or CSD reference. Only this FAA-congruent DvsHC effect on F3–F4, AVG in Study I survives control for confounding variables (t = −2.679, p=0.010).

Another significant result was observed for linear relationship between FAA and BDI score (allReg contrast) on the average referenced F3–F4 channel pair in Study V: more positive values of alpha asymmetry were associated with higher BDI scores (t = 2.489, p=0.016). This result is not consistent with the standard FAA effect, which, when calculated as right minus left alpha, should manifest as a negative correlation (negative t).

However, as we argued in the introduction, single channel analyses are not a particularly good approach to testing FAA. They may be sensitive to small changes in the topography pattern and do not provide any information about the source or physiological plausibility of the effect.

Cluster-based analyses

The next set of analyses consisted of cluster-based analyses on all frontal channel pairs (Tables 3 and 4). This approach gives a better view of the whole FAA space (correcting for multiple comparisons) especially when coupled with presentation of the effects’ topographies. We observe only one significant effect out of 30 for standard analyses (p=0.785, binomial test) and one out of 30 (p=0.785) when controlling for confounding variables.

The significant result for standard analyses corresponds to DvsHC contrast for the average referenced data in Study V (cluster p=0.026). This is the same analysis combination as the significant channel-pair result for Study V reported in the previous section. And just like in the previous section it represents a result that is not congruent with the classical FAA effect: the positive t values mean more positive right minus left FAA in depressed compared to healthy controls. We compare this effect to other analyses for DvsHC contrast in Figure 4. We do not observe any other significant effect in standard cluster-based results: neither contrasting sub-clinical vs healthy controls nor looking for a linear relationship between FAA and depression questionnaire score. Although single channels sometimes pass the significance threshold (see n significant points column in Table 3) these effects are not significant at the cluster level. In other words the clusters formed by these channels were not convincingly stronger from clusters observed under the null hypothesis (that is, when permuting the data).

Figure 4 with 4 supplements see all

Download asset Open asset

However, we observe an interesting effect of controlling for confounds. The significant result mentioned in the previous paragraph is no longer present after controlling for age, gender, and education (see Figure 5A). This change might be related to the fact that the diagnosed group has significantly lower education than the control group in Study V (t = −3.07, p=0.004, see Figure 5B), so a part of the FAA differences between the groups can be explained away by education. On the other hand, one result close to the significance threshold in the standard analyses (DvsHC, Study I, AVG, cluster p=0.069) becomes significant after controlling for gender and age (p=0.023), which may be related to the difference in age between the depressed and the control group in this study (t = 2.73, p=0.0095).

Figure 5

Download asset Open asset

Cluster-based analyses on standardized data

All previous analyses assume that asymmetry can be detected by subtracting right and left homologous channels. Because some asymmetry effects may not match such strict left vs right pattern, we performed further analyses to alleviate this issue: in this set of analyses we relied on standardization of alpha power at frontal channels instead of right minus left subtraction. Additionally, to minimize the risk of averaging out an effect confined to a narrow frequency range, we also analyzed all frequency bins in the 8–13 Hz range. This strategy is used in a set of 60 analyses (30 without and 30 with control for confounds) – we report their results in Tables 5 and 6.

Only two out of 30 analyses on standardized data without control for confounds showed statistically significant result – this is expected by chance given our alpha level (p=0.447, binomial test). Specifically, the significant results were found for: (a) DReg contrast, Study III (cluster p=0.025) and (b) DvsHC contrast, Study IV (p=0.007), both for average referenced data. Both these effects are potentially interesting, because they show a similar pattern – a positive relationship between (a) depression severity (BDI score, only the diagnosed subjects) or (b) diagnosis status and power in the higher alpha band (DReg: 11–12.5 Hz, DvsHC: 10.5–12 Hz) across many frontal channels (see Figure 6). Because it is accompanied with an inverse effect in a lower frequency band (DReg: 9–10 Hz, DvsHC: 8–9 Hz) at similar channels, averaging across the whole 8–13 Hz frequency range could lead to both effects cancelling each other in the average. Such a pattern of inverse effects across frequencies could arise due to a frequency shift of the individual alpha peak or the narrowing of the peak with depression severity but it may also be a direct consequence of standardization. However, both effects do not represent FAA as their topography is symmetrical (DReg: χ²(1)=0.153, p=0.696, DvsHC: χ²(1)=0.000, p=1) and were not replicated when using CSD reference in the same studies (Study III, Dreg: cluster p=0.259; Study IV, DvsHC: p=0.105) or performing the same contrast on data from other studies (see more: Table 5, Table 6, and Figure 6).

Figure 6 with 4 supplements see all

Download asset Open asset

Although the conclusions that can be drawn from these standardization analyses are not in favor of FAA–DD relationship, they demonstrate the strength of the proposed approach in detecting effects that might be otherwise missed when averaging across the whole alpha range or when testing only the differences on corresponding right–left channel pairs.

Source level analyses

Because observing the effect in the signal recorded from frontal channels does not guarantee that the source of this effect is frontal we conducted a second set of additional analyses using source localization with DICS beamforming. In these analyses the FAA was evaluated in the source space by subtracting power of the corresponding right and left hemisphere vertices.

Results of source level analyses are reported in Tables 7 and 8. One of the 15 source space analyses turned out statistically significant – this is expected by chance given our alpha value (p=0.537, binomial test). The same number of significant results was found when controlling for confounding variables (1/15, p=0.537).

The significant effect was found for DvsHC contrast in Study IV (p=0.010, see Figure 7) and remains significant when controlling for confounds (cluster p=0.011; see Table 8). It represents more negative FAA values for depressed compared to healthy individuals, which is congruent with the traditional FAA effect.

Figure 7 with 4 supplements see all

Download asset Open asset

In most of the analyses the pattern and sign of the t values points towards a more left-sided effect. For example in the allReg contrast: the negative t values suggest lower R–L differences in high than in low BDI participants, which means more left-sided alpha power with higher BDI. Although this pattern seems to be in line with FAA–DD literature, almost all of the source space effects are weak and do not survive the correction for multiple comparisons. However, it is important to remember that individual MRI scans and channel locations were not available in the present study: their availability would lead to lower error in source reconstruction.

Analyses on aggregated data

Finally, to overcome the relatively low statistical power of analyses on separate data sets we aggregate data from all studies that include identical contrasts and perform analyses on the aggregated data. Before aggregation we tested whether the FAA values from both selected channel pairs have similar scale across the five studies with a Levene test. Because the scale was significantly different across studies (F3–F4: W = 8.68, p<0.0001; F7–F8: W = 5.21, p=0.002) and because such scale differences can arise from lab-specific equipment or adopted impedance threshold, we z-scored the FAA values within each study before aggregation. All aggregated channel pair analyses can be seen in Figure 8, Figure 8—figure supplement 1, and Table 9. For brevity we discuss only the results for DvsC and allReg contrasts for average referenced channel pairs.

Figure 8 with 2 supplements see all

Download asset Open asset

Table 9

No.	Contrast	Space	N	Control for confounds		Aggregated channel pair analyses
						Pair 1 (F3–F4)				Pair 2 (F7–F8)
						ES	CI	BF01		ES	CI	BF01
1	DvsHC	avg	102 vs 144	−		0.147	[−0.111, 0.396]	3.831		0.098	[−0.161, 0.338]	5.405
2	DvsHC	avg	102 vs 143	+		−0.011	[−0.264, 0.244]	7.042		−0.006	[−0.266, 0.240]	7.042
3	DvsHC	csd	102 vs 144	−		0.188	[−0.069, 0.449]	2.597		−0.100	[−0.354, 0.164]	5.319
4	DvsHC	csd	102 vs 143	+		0.103	[−0.159, 0.362]	5.236		−0.164	[−0.417, 0.108]	3.300
5	SvsHC	avg	77 vs 121	−		0.065	[−0.236, 0.359]	5.780		−0.025	[−0.320, 0.240]	6.250
6	SvsHC	avg	77 vs 120	+		0.024	[−0.269, 0.315]	6.211		−0.144	[−0.447, 0.140]	4.016
7	SvsHC	csd	77 vs 121	−		0.053	[−0.223, 0.355]	5.952		−0.108	[−0.372, 0.179]	4.878
8	SvsHC	csd	77 vs 120	+		0.053	[−0.222, 0.350]	5.952		−0.110	[−0.389, 0.179]	4.831
9	DReg	avg	102	−		0.188	[0.006, 0.354]	1.387		0.007	[−0.186, 0.188]	8.065
10	DReg	avg	102	+		0.175	[−0.018, 0.348]	1.742		−0.029	[−0.211, 0.155]	7.752
11	DReg	csd	102	−		0.099	[−0.072, 0.260]	4.975		0.065	[−0.136, 0.252]	6.579
12	DReg	csd	102	+		0.051	[−0.131, 0.216]	7.092		0.048	[−0.157, 0.255]	7.194
13	allReg	avg	315	−		0.085	[−0.017, 0.184]	4.651		0.029	[−0.080, 0.126]	12.5
14	allReg	avg	314	+		0.041	[−0.063, 0.143]	10.87		0.001	[−0.120, 0.104]	14.085
15	allReg	csd	315	−		0.059	[−0.054, 0.166]	8.333		−0.026	[−0.141, 0.082]	12.821
16	allReg	csd	314	+		0.055	[−0.056, 0.168]	8.772		−0.048	[−0.158, 0.059]	9.901

1. DvsHC – diagnosed and healthy controls; SvsHC – sub-clinical and healthy controls; allReg – linear regression for all subjects together; DReg – linear regression for only diagnosed subjects; avg – average reference; csd – current source density.

Aggregated DvsHC contrast analysis encompasses 246 participants (102 diagnosed and 144 healthy controls) and 245 when controlling for confounds (one participant from the control group was removed due to missing information on confounding variables). For the F3–F4 channel pair the Cohen’s d is 0.147 (CI = [−0.111, 0.396]) and −0.011 when controlling for confounds (CI = [−0.264, 0.244]). Both confidence intervals exclude all but small effect sizes (classical FAA effect for right–left, diagnosed – controls should be negative). The effect sizes for F7–F8 channel pair are similar: d = 0.098, CI = [−0.161, 0.338] without control for confounds and d = −0.006, CI = [−0.266, 0.240] when controlling for confounding variables.

To quantify the support for the null hypothesis we calculate Bayes factors for the null (BF01). For F3–F4 channel pair the BF01 equals 3.831, which means that the data are almost four times more likely under the null than alternative hypothesis. When controlling for confounding variables the BF01 increases to 7.042. For F7–F8 channel pair the BF01 are: 5.405 and 7.042 when controlling for confounds. Bayes factors between 3 and 10 are considered moderate evidence, so the results provide moderate evidence for no FAA difference between diagnosed and healthy individuals.

Aggregated allReg contrast analysis includes 315 participants (314 when controlling for confounds). For F3–F4 channel pair the Pearson’s r is 0.085 (CI = [−0.017, 0.184]) and decreases to 0.041 when controlling for confounds (CI = [−0.063, 0.143]). Corresponding Bayes factors for the null are: 4.651 and 10.87 which suggests moderate to strong evidence for no relationship between FAA and depression score.

We also aggregate the data from relevant studies in the source space and then perform cluster-based permutation tests for all defined contrasts. Just like in the aggregated channel pair analyses, before aggregation the data are z-scored within each study (each source vertex is z-scored across participants separately) to avoid creating non-normal distributions by joining data of different scale. The results of the aggregated source space analyses can be found in Table 10.

Table 10

No.	Contrast	N	Control for confounds	Aggregated source level analyses
				Min t	Max t	n significant points	n clusters	Largest cluster size	Largest cluster p
1	DvsHC	102 vs 144	−	−2.719	0.956	178	5	100	0.077
2	DvsHC	102 vs 143	+	−2.846	0.908	211	5	116	0.066
3	SvsHC	77 vs 121	−	−1.831	−0.11	0	0	NA	NA
4	SvsHC	77 vs 120	+	−1.871	0.182	0	0	NA	NA
5	DReg	102	−	−2.387	1.376	31	3	25	0.328
6	DReg	102	+	−3.095	1.354	112	3	68	0.155
7	allReg	315	−	−3.044	0.411	174	6	160	0.067
8	allReg	314	+	−3.107	0.282	235	6	195	0.054

1. DvsHC – diagnosed and healthy controls; SvsHC – sub-clinical and healthy controls; allReg – linear regression for all subjects together; DReg – linear regression for only diagnosed subjects.

Although no analysis yields nominally significant results it is interesting to note that contrasts DvsHC and allReg are at the conventional ‘trend’ level (DvsHC: 0.077 and 0.066 when controlling for confounds; allReg: 0.067 and 0.054 when controlling for confounds) and their direction agrees with the traditional FAA effect.

We conducted a multiverse analysis of EEG data sets from five independent studies, with 388 participants and 270 analyses in total, to test the robustness and credibility of the relationship between FAA and depressive mood. We performed 120 replicatory single channel pairs analyses and 60 corresponding cluster-based analyses. We have also conducted 90 additional analyses addressing some of the limitations of current FAA studies. Out of 270 performed analyses only 13 produced statistically significant results – a result expected by chance when using 0.05 alpha level (binomial test, p=0.595). Moreover, more than half of the significant results (8/13) are incongruent with the traditional FAA effect: by either showing the opposite direction of the effect (6) or not showing asymmetry (two significant effects on standardized data). Overall, the conducted analyses do not provide a basis to reject the null hypothesis of no relationship between resting state FAA and DD.

Our conclusion is similar to this formulated by other research groups (Kaiser et al., 2018b; van der Vinne et al., 2017), stating that treating FAA as a biomarker of DD is not sufficiently empirically grounded. As our data shows, this skepticism is not limited to single channel pair analyses – improving on the limitations of methods commonly used in FAA literature does not change the pattern of results.

Despite this, FAA is one of the most common indicators of DD with a long history of successful studies – it might be difficult to believe that all previous FAA research represents Type I errors. Therefore it is worth considering that we just fail to detect this effect here. First, the FAA effect size may be too small to be reliably observed with a small to moderate sample size. The average number of participants per study is 78 in our case, but most analyses contain around 25 participants per group, which grants sufficient power to detect mostly large effects. Although single analyses reported here can be deemed inconclusive, the whole multiverse set of analyses is incompatible with the presence of moderate to strong relationship between FAA and DD. To strengthen this point we performed analyses on data aggregated across studies (see section Analyses on aggregated data) showing estimated effect sizes, their confidence intervals, and Bayes factors for the null hypothesis (see Figure 8 and Figure 8—figure supplement 1). All confidence intervals for diagnosed vs healthy controls group contrast (DvsHC) and linear relationship between FAA and diagnosis score (allReg) exclude strong and moderate effects in the direction compatible with traditional FAA effect. Moreover, Bayes factors indicate that there is moderate evidence for the null hypothesis (or moderate to strong evidence for the null in allReg contrast). Although we cannot exclude a small FAA–DD relationship, if the effect was in this range then most published studies would have been underpowered to detect it. This line of thought is also supported in the meta-analysis by van der Vinne et al., 2017, which shows that studies with larger samples were less likely to report high effect sizes. For example, the largest EEG FAA study on a sample of 1008 DD patients and 336 controls did not confirm the diagnostic value of FAA in DD (Arns et al., 2016). Such pattern of results suggests publication bias or that the FAA–DD effect, if it exists, is detectable only in highly selected samples and is of small magnitude on the population level.

Although the collected studies contain data from sub-clinical, mild, as well as major depression patients, we do not think our results can be explained by the level of depression severity. Clinical participants in Studies I, III, IV, and V and sub-clinical participants in Studies II and III manifested a wide range of DD symptoms indicated by BDI/PHQ-9 scores (see Figure 1 for BDI histograms), but regression analysis (DReg) did not reveal any relationship between FAA and BDI score in the aggregated analyses. This means that participants with stronger DD symptoms were not better characterized by a specific FAA pattern even when we controlled for confounding variables like age, gender, and education. However, given that 56 out of 102 participants (54.9%) in the aggregated clinical group were diagnosed with mild DD, it would be interesting to repeat the analyses presented here on a data set with more major DD patients.

Smith et al., 2017 previously suggested that the relationship between FAA and DD is stronger when the participant is given some emotion-related task, as opposed to resting condition, where the task is unspecified. Although this is possible, we wanted to stay true to the design of most FAA–DD studies, which measure EEG during rest. An interesting approach for future studies would be to compare rest blocks separated by an emotional task (see, for example, Beeney et al., 2014).

CSD has been previously recommended in the literature (Kayser and Tenke, 2015; Stewart et al., 2014) for studies on FAA because it reduces volume conduction and makes topographies more focal. We do not see support for the claim that CSD is more sensitive to FAA in our results. We even think it may be the contrary – almost all significant or trend-level effects disappear with CSD reference. The fact that CSD produces more focal topographies coupled with potential high variability of topographies across subjects may result in lower probability of detecting an effect. Given that the orientation of the sources responsible for the effect of interest may be variable across subjects, increasing the focality of their projections may only exacerbate the issue.

In contrast to CSD, we see some indication of the traditional FAA effect in the source space analyses. Although only 2/30 source space analyses are significant, partitioning the effects into contrasts (see Table 11) reveals 2/8 significant DvsHC results, which is only barely consistent with 5% error rate (p=0.057, binomial test). Also the direction of the effect in almost all source results is consistent with the traditional FAA effect (negative t values). Finally, the same consistency of the effect direction can be observed in the source space analyses on aggregated data (see Table 10, Figure 9, and Figure 9—figure supplement 1) – all source space vertices with an uncorrected significant effect show negative effect direction. The DvsHC and allReg contrasts on the aggregated data are at the trend level (p=0.077 and p=0.067, respectively) and it might seem that using more rigorous source localization (individual MRI scans and channel positions) could lead to a significant effect. Nevertheless it is difficult to speculate based on statistical tendency – with more subjects or more precise source localization the results might equally likely land further away from the alpha threshold.

Figure 9 with 2 supplements see all

Download asset Open asset

Previous studies have suggested that FAA effect may manifest differently depending on gender (Jaworska et al., 2012; Stewart et al., 2010; van der Vinne et al., 2017). Because all the analyses with control for confounding variables that we have conducted control only for the main effect of gender, we additionally tested the presence of an interaction between gender and diagnosis (or depression score) in predicting FAA. This set of analyses were restricted only to the aggregated data with control for confounds. None of these analyses demonstrated a significant interaction effect (see Figure 8—figure supplement 2 and Figure 9—figure supplement 2).

A challenge for future FAA studies would be to move beyond a ‘marker-only’ approach, describe the theoretical assumptions behind FAA in more detail, and let these assumptions dictate an adequate analytical approach. For example, the assumption that FAA is a phenomenon with a source in the frontal cortex is impossible to address using only a few frontal channel pairs. Using source localization (like DICS Beamforming used here) or source separation (for example, spatial filtering with generalized eigendecomposition/common spatial pattern [Koles et al., 1990; Parra and Sajda, 2003; Tomé, 2006]; or SPACE decomposition [van der Meij et al., 2015]; Roemer [van der Meij et al., 2016]) should be preferred when looking for the answer to this question. This point is important because frontal alpha sources are rarely measured reliably with EEG at the channel level: strong occipital and parietal alpha sources dominate alpha power recorded at frontal channels. As a result measuring FAA at the channel level could lead to poor signal to noise ratio and consequently to small effect size and low probability to observe a true FAA–DD relationship. On the other hand, if FAA does not originate in the frontal regions, it should be measured and interpreted differently. Such scenario is not unlikely because frontal alpha sources are generally difficult to detect with EEG/MEG. For example, Roemer van der Meij et al., 2016, using an advanced source separation method, found that 86.6% of the alpha components detected across subjects were occipito-parietal and only 1% (4/380) were frontal. If frontal alpha sources are difficult to detect using a source separation method designed to capture oscillatory sources, then it is likely that these sources are rarely observed at the channel level at all.

If FAA is not of frontal origin, then where could it come from? This is an open empirical question and we can only offer speculation here. Jiang et al., 2016 have shown that the power of posterior alpha oscillations is reduced in depressed individuals and that this reduction strongly correlates with depression severity. Assuming that frontal projections from occipital or parietal sources will not be perfectly symmetrical, a difference metric like FAA may be sensitive to posterior alpha power. Another possibility is that FAA originates from asymmetry at the source level; Smith et al., 2018 demonstrated that channel-level FAA is related to source level asymmetry in frontal motor regions. However, the authors did not look for correlations with FAA in the full source space, but restricted their analyses to the R–L source level asymmetry and in consequence their analysis is insensitive to symmetrical sources of FAA. Our data and analyses cannot be conclusive in this regard because without a strong and reliable channel-level FAA–DD effect it is difficult to look for its source level correlate.

Tackling all the mentioned issues would require to systemize and unify the FAA methodology for the benefit of future studies. So far, Kaiser et al., 2018a proposed guidelines for methodology regarding subjects selection and controlling for confounding variables. Smith et al., 2017 also suggested possible improvements in experimental procedures and EEG signal preprocessing. We believe there is still room for improvement in the signal analysis standards of FAA studies. Below we summarize our arguments and propose additional guidelines for EEG data analysis in FAA research:

• Always show the topography of the effects. Lack of topographical plots hinders interpretation in terms of both potential neural origin of the effect and its physiological reliability. It is a good idea to also add topographical plots of group averages: both for alpha power and alpha asymmetry (see Figure 4—figure supplement 4). Such visualizations can clarify the studied effect: when FAA is calculated as a R–L difference and is compared between groups, reasoning about difference between differences ((R–L) − (R–L)) can be unnecessarily complex.

• Conduct analysis on all frontal electrodes (or even all available electrodes) with correction for multiple comparisons. We recommend using the cluster-based permutation test (Maris and Oostenveld, 2007), as it is versatile and implemented in multiple software packages: mne-python (Gramfort et al., 2013; Gramfort et al., 2014) and fieldtrip (Oostenveld et al., 2011) for example, but the fieldtrip implementation is available also through EEGLAB (Delorme and Makeig, 2004) and brainstorm (Tadel et al., 2011).

• Try not to restrict the analysis to left minus right subtraction on averaged frequencies. As we show in the analyses on standardized data, avoiding subtraction and frequency averaging can uncover interesting effects that could otherwise be missed. Extending the search space to frequencies is straightforward when using the cluster-based permutation test.

• Perform analysis in the source-space if possible. Source localization allows to estimate the source of the signal more reliably and obtain a better signal to noise ratio (see, for example, van Es and Schoffelen, 2019). However, to minimize source localization error individual MRI scans are required. Other methods focusing on source-separation like ICA, GED, or SPACE allow one to disentangle signal contributions from independent sources and increase signal to noise ratio. CSD was also proposed in this context in FAA literature before, but although it can mitigate some of the issues arising from volume conduction, it does not provide source localization or separation.

• Do not restrict the analysis to group contrasts if linear predictors are available. Using linear regression allows to take covariates into account and test hypotheses in a more detailed manner.

We included five data sets in the analyses. These data sets were obtained in five independent studies: Studies I–III have been collected by the authors of the article; Studies IV and V are publicly available: the data from Study IV were obtained from the PREDiCT repository (Cavanagh et al., 2017), while Study V data come from the MODMA database (Cai et al., 2020).

EEG data has been deposited to Dryad, and can be found under the https://doi.org/10.5061/dryad.5x69p8d18.

1. 1. Aleksandra K
   2. Mikołaj M
   3. Anastasia R
   4. Aneta B

   (2020) Dryad Digital Repository

   No relationship between frontal alpha asymmetry and depressive disorders in a multiverse analysis of five studies.

   https://doi.org/10.5061/dryad.5x69p8d18

1. 1. Cavanagh JF
   2. Allen JB

   (2017) PREDiCT

   ID d006. Depression Rest.

   http://predict.cs.unm.edu/downloads.php
2. 1. Cai H
   2. Gao Y
   3. Sun S
   4. Li N
   5. Tian F
   6. Xiao H
   7. Li J
   8. Yang Z
   9. Li X
   10. Zhao Q
   11. Liu Z
   12. Yao Z
   13. Yang M
   14. Peng H
   15. Zhu J
   16. Zhang X
   17. Gao G
   18. Zheng F
   19. Li R
   20. Guo Z
   21. Ma R
   22. Yang J
   23. Zhang L
   24. Hu X
   25. Li Y
   26. Hu B

   (2015) MODMA

   ID EEG_128channels_resting_lanzhou_2015. EEG_128channels_resting_lanzhou_2015.

   http://modma.lzu.edu.cn/data/application/

1. 1. Allen JJ
   2. Urry HL
   3. Hitt SK
   4. Coan JA

   (2004) The stability of resting frontal electroencephalographic asymmetry in depression

   Psychophysiology 41:269–280.

   https://doi.org/10.1111/j.1469-8986.2003.00149.x

   • PubMed
   • Google Scholar
2. 1. Arns M
   2. Bruder G
   3. Hegerl U
   4. Spooner C
   5. Palmer DM
   6. Etkin A
   7. Fallahpour K
   8. Gatt JM
   9. Hirshberg L
   10. Gordon E

   (2016) EEG alpha asymmetry as a gender-specific predictor of outcome to acute treatment with different antidepressant medications in the randomized iSPOT-D study

   Clinical Neurophysiology 127:509–519.

   https://doi.org/10.1016/j.clinph.2015.05.032

   • PubMed
   • Google Scholar
3. 1. Baskaran A
   2. Milev R
   3. McIntyre RS

   (2012) The neurobiology of the EEG biomarker as a predictor of treatment response in depression

   Neuropharmacology 63:507–513.

   https://doi.org/10.1016/j.neuropharm.2012.04.021

   • PubMed
   • Google Scholar
4. 1. Beck AT
   2. Ward CH
   3. Mendelson M
   4. Mock J
   5. Erbaugh J

   (1961) An inventory for measuring depression

   Archives of General Psychiatry 4:561–571.

   https://doi.org/10.1001/archpsyc.1961.01710120031004

   • PubMed
   • Google Scholar
5. Book

   1. Beck A
   2. Steer R
   3. Brown G

   (1996)

   Manual for the Beck Depression Inventory-II (BDI-II)

   San Antonio, TX: Psychological Corporation.

   • Google Scholar
6. 1. Beeney JE
   2. Levy KN
   3. Gatzke-Kopp LM
   4. Hallquist MN

   (2014) EEG asymmetry in borderline personality disorder and depression following rejection

   Personality Disorders: Theory, Research, and Treatment 5:178–185.

   https://doi.org/10.1037/per0000032

   • PubMed
   • Google Scholar
7. Preprint

   1. Botvinik-Nezer R
   2. Holzmeister F
   3. Camerer CF
   4. Dreber A
   5. Huber J
   6. Johannesson M
   7. Kirchler M
   8. Iwanir R
   9. Mumford JA
   10. Adcock A

   (2019) Variability in the analysis of a single neuroimaging dataset by many teams

   bioRxiv.

   https://doi.org/10.1101/843193

   • Google Scholar
8. Preprint

   1. Cai H
   2. Gao Y
   3. Sun S
   4. Li N
   5. Tian F
   6. Xiao H
   7. Li J

   (2020) MODMA dataset: a Multi-model open dataset for Mental-disorder analysis

   arXiv.

   https://arxiv.org/abs/2002.09283

   • Google Scholar
9. 1. Carvalho A
   2. Moraes H
   3. Silveira H
   4. Ribeiro P
   5. Piedade RA
   6. Deslandes AC
   7. Laks J
   8. Versiani M

   (2011) EEG frontal asymmetry in the depressed and remitted elderly: is it related to the trait or to the state of depression?

   Journal of Affective Disorders 129:143–148.

   https://doi.org/10.1016/j.jad.2010.08.023

   • PubMed
   • Google Scholar
10. 1. Cavanagh JF
    2. Bismark AJ
    3. Frank MJ
    4. Allen JJ

    (2011) Larger error signals in major depression are associated with better avoidance learning

    Frontiers in Psychology 2:331.

    https://doi.org/10.3389/fpsyg.2011.00331

    • PubMed
    • Google Scholar
11. 1. Cavanagh JF
    2. Napolitano A
    3. Wu C
    4. Mueen A

    (2017) The patient repository for EEG data + computational tools (PRED+CT)

    Frontiers in Neuroinformatics 11:67.

    https://doi.org/10.3389/fninf.2017.00067

    • PubMed
    • Google Scholar
12. 1. Cavanagh JF
    2. Bismark AW
    3. Frank MJ
    4. Allen JJB

    (2019) Multiple dissociations between comorbid depression and anxiety on reward and Punishment Processing: Evidence From Computationally Informed EEG

    Computational Psychiatry 3:1–17.

    https://doi.org/10.1162/CPSY_a_00024

    • Google Scholar
13. 1. Cohen MX

    (2015) Comparison of different spatial transformations applied to EEG data: a case study of error processing

    International Journal of Psychophysiology 97:245–257.

    https://doi.org/10.1016/j.ijpsycho.2014.09.013

    • PubMed
    • Google Scholar
14. 1. Cohen MX
    2. Gulbinaite R

    (2014) Five methodological challenges in cognitive electrophysiology

    NeuroImage 85 Pt 2:702–710.

    https://doi.org/10.1016/j.neuroimage.2013.08.010

    • PubMed
    • Google Scholar
15. 1. Dale AM
    2. Fischl B
    3. Sereno MI

    (1999) Cortical surface-based analysis. I. segmentation and surface reconstruction

    NeuroImage 9:179–194.

    https://doi.org/10.1006/nimg.1998.0395

    • PubMed
    • Google Scholar
16. Book

    1. Davidson RJ

    (1984)

    Affect, Cognition, and Hemispheric Specialization

    In: Izard CE, Kagan J, Zajonc R, editors. Emotion, Cognition, and Behavior. New York: Cambridge University Press. pp. 320–365.

    • Google Scholar
17. 1. Davidson RJ

    (2004) What does the prefrontal cortex "do" in affect: perspectives on frontal EEG asymmetry research

    Biological Psychology 67:219–234.

    https://doi.org/10.1016/j.biopsycho.2004.03.008

    • PubMed
    • Google Scholar
18. 1. de Aguiar Neto FS
    2. Rosa JLG

    (2019) Depression biomarkers using non-invasive EEG: a review

    Neuroscience & Biobehavioral Reviews 105:83–93.

    https://doi.org/10.1016/j.neubiorev.2019.07.021

    • PubMed
    • Google Scholar
19. 1. De Raedt R
    2. Franck E
    3. Fannes K
    4. Verstraeten E

    (2008) Is the relationship between frontal EEG alpha asymmetry and depression mediated by implicit or explicit self-esteem?

    Biological Psychology 77:89–92.

    https://doi.org/10.1016/j.biopsycho.2007.06.004

    • PubMed
    • Google Scholar
20. 1. Delacre M
    2. Lakens D
    3. Leys C

    (2017) Why psychologists should by default use Welch's t-test Instead of Student's t-test

    International Review of Social Psychology 30:92–101.

    https://doi.org/10.5334/irsp.82

    • Google Scholar
21. 1. Deldin PJ
    2. Chiu P

    (2005) Cognitive restructuring and EEG in major depression

    Biological Psychology 70:141–151.

    https://doi.org/10.1016/j.biopsycho.2005.01.003

    • PubMed
    • Google Scholar
22. 1. Delorme A
    2. Makeig S

    (2004) EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics including independent component analysis

    Journal of Neuroscience Methods 134:9–21.

    https://doi.org/10.1016/j.jneumeth.2003.10.009

    • PubMed
    • Google Scholar
23. 1. Fischl B
    2. Sereno MI
    3. Tootell RB
    4. Dale AM

    (1999) High-resolution intersubject averaging and a coordinate system for the cortical surface

    Human Brain Mapping 8:272–284.

    https://doi.org/10.1002/(SICI)1097-0193(1999)8:4<272::AID-HBM10>3.0.CO;2-4

    • PubMed
    • Google Scholar
24. 1. Gelman A
    2. Loken E

    (2014) The statistical crisis in science

    American Scientist 102:460.

    https://doi.org/10.1511/2014.111.460

    • Google Scholar
25. 1. Gold C
    2. Fachner J
    3. Erkkilä J

    (2013) Validity and reliability of electroencephalographic frontal alpha asymmetry and frontal midline theta as biomarkers for depression

    Scandinavian Journal of Psychology 54:118–126.

    https://doi.org/10.1111/sjop.12022

    • Google Scholar
26. 1. Gramfort A
    2. Luessi M
    3. Larson E
    4. Engemann DA
    5. Strohmeier D
    6. Brodbeck C
    7. Goj R
    8. Jas M
    9. Brooks T
    10. Parkkonen L
    11. Hämäläinen M

    (2013) MEG and EEG data analysis with MNE-Python

    Frontiers in Neuroscience 7:267.

    https://doi.org/10.3389/fnins.2013.00267

    • PubMed
    • Google Scholar
27. 1. Gramfort A
    2. Luessi M
    3. Larson E
    4. Engemann DA
    5. Strohmeier D
    6. Brodbeck C
    7. Parkkonen L
    8. Hämäläinen MS

    (2014) MNE software for processing MEG and EEG data

    NeuroImage 86:446–460.

    https://doi.org/10.1016/j.neuroimage.2013.10.027

    • PubMed
    • Google Scholar
28. 1. Greve DN
    2. Van der Haegen L
    3. Cai Q
    4. Stufflebeam S
    5. Sabuncu MR
    6. Fischl B
    7. Brysbaert M

    (2013) A surface-based analysis of language lateralization and cortical asymmetry

    Journal of Cognitive Neuroscience 25:1477–1492.

    https://doi.org/10.1162/jocn_a_00405

    • PubMed
    • Google Scholar
29. 1. Gross J
    2. Kujala J
    3. Hamalainen M
    4. Timmermann L
    5. Schnitzler A
    6. Salmelin R

    (2001) Dynamic imaging of coherent sources: studying neural interactions in the human brain

    PNAS 98:694–699.

    https://doi.org/10.1073/pnas.98.2.694

    • PubMed
    • Google Scholar
30. 1. Hipp JF
    2. Siegel M

    (2013) Dissociating neuronal gamma-band activity from cranial and ocular muscle activity in EEG

    Frontiers in Human Neuroscience 7:7.

    https://doi.org/10.3389/fnhum.2013.00338

    • PubMed
    • Google Scholar
31. 1. Ho J
    2. Tumkaya T
    3. Aryal S
    4. Choi H
    5. Claridge-Chang A

    (2019) Moving beyond P values: data analysis with estimation graphics

    Nature Methods 16:565–566.

    https://doi.org/10.1038/s41592-019-0470-3

    • PubMed
    • Google Scholar
32. 1. Imperatori C
    2. Farina B
    3. Valenti EM
    4. Di Poce A
    5. D'Ari S
    6. De Rossi E
    7. Murgia C
    8. Carbone GA
    9. Massullo C
    10. Della Marca G

    (2019) Is resting state frontal alpha connectivity asymmetry a useful index to assess depressive symptoms? A preliminary investigation in a sample of university students

    Journal of Affective Disorders 257:152–159.

    https://doi.org/10.1016/j.jad.2019.07.034

    • PubMed
    • Google Scholar
33. 1. Iosifescu DV
    2. Greenwald S
    3. Devlin P
    4. Mischoulon D
    5. Denninger JW
    6. Alpert JE
    7. Fava M

    (2009) Frontal EEG predictors of treatment outcome in major depressive disorder

    European Neuropsychopharmacology 19:772–777.

    https://doi.org/10.1016/j.euroneuro.2009.06.001

    • PubMed
    • Google Scholar
34. 1. Jaworska N
    2. Blier P
    3. Fusee W
    4. Knott V

    (2012) ?? power, α asymmetry and anterior cingulate cortex activity in depressed males and females

    Journal of Psychiatric Research 46:1483–1491.

    https://doi.org/10.1016/j.jpsychires.2012.08.003

    • PubMed
    • Google Scholar
35. 1. Jesulola E
    2. Sharpley CF
    3. Agnew LL

    (2017) The effects of gender and depression severity on the association between alpha asymmetry and depression across four brain regions

    Behavioural Brain Research 321:232–239.

    https://doi.org/10.1016/j.bbr.2016.12.035

    • PubMed
    • Google Scholar
36. 1. Jiang H
    2. Popov T
    3. Jylänki P
    4. Bi K
    5. Yao Z
    6. Lu Q
    7. Jensen O
    8. van Gerven MA

    (2016) Predictability of depression severity based on posterior alpha oscillations

    Clinical Neurophysiology 127:2108–2114.

    https://doi.org/10.1016/j.clinph.2015.12.018

    • PubMed
    • Google Scholar
37. 1. Kaiser AK
    2. Doppelmayr M
    3. Iglseder B

    (2018a) Electroencephalogram alpha asymmetry in geriatric depression : valid or vanished?

    Zeitschrift Für Gerontologie Und Geriatrie 51:200–205.

    https://doi.org/10.1007/s00391-016-1108-z

    • PubMed
    • Google Scholar
38. 1. Kaiser AK
    2. Gnjezda MT
    3. Knasmüller S
    4. Aichhorn W

    (2018b) Electroencephalogram alpha asymmetry in patients with depressive disorders: current perspectives

    Neuropsychiatric Disease and Treatment 14:1493–1504.

    https://doi.org/10.2147/NDT.S137776

    • PubMed
    • Google Scholar
39. 1. Kayser J
    2. Tenke CE

    (2015) On the benefits of using surface laplacian (current source density) methodology in electrophysiology

    International Journal of Psychophysiology 97:171–173.

    https://doi.org/10.1016/j.ijpsycho.2015.06.001

    • PubMed
    • Google Scholar
40. 1. Kemp AH
    2. Griffiths K
    3. Felmingham KL
    4. Shankman SA
    5. Drinkenburg W
    6. Arns M
    7. Clark CR
    8. Bryant RA

    (2010) Disorder specificity despite comorbidity: resting EEG alpha asymmetry in major depressive disorder and post-traumatic stress disorder

    Biological Psychology 85:350–354.

    https://doi.org/10.1016/j.biopsycho.2010.08.001

    • PubMed
    • Google Scholar
41. 1. Kentgen LM
    2. Tenke CE
    3. Pine DS
    4. Fong R
    5. Klein RG
    6. Bruder GE

    (2000) Electroencephalographic asymmetries in adolescents with major depression: influence of comorbidity with anxiety disorders

    Journal of Abnormal Psychology 109:797–802.

    https://doi.org/10.1037/0021-843X.109.4.797

    • Google Scholar
42. 1. Knott V
    2. Mahoney C
    3. Kennedy S
    4. Evans K

    (2001) EEG power, frequency, asymmetry and coherence in male depression

    Psychiatry Research: Neuroimaging 106:123–140.

    https://doi.org/10.1016/S0925-4927(00)00080-9

    • PubMed
    • Google Scholar
43. 1. Koles ZJ
    2. Lazar MS
    3. Zhou SZ

    (1990) Spatial patterns underlying population differences in the background EEG

    Brain Topography 2:275–284.

    https://doi.org/10.1007/BF01129656

    • PubMed
    • Google Scholar
44. 1. Kroenke K
    2. Spitzer RL
    3. Williams JB

    (2001) The PHQ-9: validity of a brief depression severity measure

    Journal of General Internal Medicine 16:606–613.

    https://doi.org/10.1046/j.1525-1497.2001.016009606.x

    • PubMed
    • Google Scholar
45. 1. Li X
    2. Jing Z
    3. Hu B
    4. Zhu J
    5. Zhong N
    6. Li M
    7. Ding Z
    8. Yang J
    9. Zhang L
    10. Feng L
    11. Majoe D

    (2017) A Resting-State brain functional network study in MDD based on minimum spanning tree analysis and the hierarchical clustering

    Complexity 2017:1–11.

    https://doi.org/10.1155/2017/9514369

    • Google Scholar
46. 1. Lubar JF
    2. Congedo M
    3. Askew JH

    (2003) Low-resolution electromagnetic tomography (LORETA) of cerebral activity in chronic depressive disorder

    International Journal of Psychophysiology 49:175–185.

    https://doi.org/10.1016/S0167-8760(03)00115-6

    • Google Scholar
47. Software

    1. Magnuski M

    (2020a) GitHub repository: borsar. Borsar

    GitHub repository: borsar. Borsar.

    https://github.com/mmagnuski/borsar
48. Software

    1. Magnuski M

    (2020b) GitHub repository: eegDb. eegDb

    GitHub repository: eegDb. eegDb.

    https://github.com/mmagnuski/eegDb
49. Software

    1. Magnuski M

    (2020c) GitHub repository: sarna. Sarna

    GitHub repository: sarna. Sarna.

    https://github.com/mmagnuski/sarna
50. Software

    1. Magnuski M
    2. Ruban A

    (2020) GitHub repository: DiamSar. DiamSar

    GitHub repository: DiamSar. DiamSar.

    https://github.com/mmagnuski/DiamSar
51. 1. Maris E
    2. Oostenveld R

    (2007) Nonparametric statistical testing of EEG- and MEG-data

    Journal of Neuroscience Methods 164:177–190.

    https://doi.org/10.1016/j.jneumeth.2007.03.024

    • PubMed
    • Google Scholar
52. 1. Mathersul D
    2. Williams LM
    3. Hopkinson PJ
    4. Kemp AH

    (2008) Investigating models of affect: relationships among EEG alpha asymmetry, depression, and anxiety

    Emotion 8:560–572.

    https://doi.org/10.1037/a0012811

    • PubMed
    • Google Scholar
53. 1. McFarland MJ
    2. Wagner BG

    (2015) Does a college education reduce depressive symptoms in american young adults?

    Social Science & Medicine 146:75–84.

    https://doi.org/10.1016/j.socscimed.2015.09.029

    • Google Scholar
54. 1. McMenamin BW
    2. Shackman AJ
    3. Maxwell JS
    4. Bachhuber DR
    5. Koppenhaver AM
    6. Greischar LL
    7. Davidson RJ

    (2010) Validation of ICA-based myogenic artifact correction for scalp and source-localized EEG

    NeuroImage 49:2416–2432.

    https://doi.org/10.1016/j.neuroimage.2009.10.010

    • PubMed
    • Google Scholar
55. 1. Nolen-Hoeksema S

    (2001) Gender differences in depression

    Current Directions in Psychological Science 10:173–176.

    https://doi.org/10.1111/1467-8721.00142

    • Google Scholar
56. 1. Oostenveld R
    2. Fries P
    3. Maris E
    4. Schoffelen JM

    (2011) FieldTrip: open source software for advanced analysis of MEG, EEG, and invasive electrophysiological data

    Computational Intelligence and Neuroscience 2011:1–9.

    https://doi.org/10.1155/2011/156869

    • PubMed
    • Google Scholar
57. 1. Parameshwaran D
    2. Thiagarajan TC

    (2019) Modernization, wealth and the emergence of strong alpha oscillations in the human EEG

    Cold Spring Harbor Laboratory 1:125898.

    https://doi.org/10.1101/125898

    • Google Scholar
58. 1. Parra L
    2. Sajda P

    (2003) Blind source separation via generalized eigenvalue decomposition

    Journal of Machine Learning Research : JMLR 4:1261–1269.

    https://doi.org/10.1162/jmlr.2003.4.7-8.1261

    • Google Scholar
59. 1. Rouder JN
    2. Speckman PL
    3. Sun D
    4. Morey RD
    5. Iverson G

    (2009) Bayesian t tests for accepting and rejecting the null hypothesis

    Psychonomic Bulletin & Review 16:225–237.

    https://doi.org/10.3758/PBR.16.2.225

    • PubMed
    • Google Scholar
60. 1. Schaffer CE
    2. Davidson RJ
    3. Saron C

    (1983)

    Frontal and parietal electroencephalogram asymmetry in depressed and nondepressed subjects

    Biological Psychiatry 18:753–762.

    • PubMed
    • Google Scholar
61. 1. Shackman AJ
    2. McMenamin BW
    3. Slagter HA
    4. Maxwell JS
    5. Greischar LL
    6. Davidson RJ

    (2009) Electromyogenic artifacts and electroencephalographic inferences

    Brain Topography 22:7–12.

    https://doi.org/10.1007/s10548-009-0079-4

    • PubMed
    • Google Scholar
62. 1. Sheehan DV
    2. Lecrubier Y
    3. Sheehan KH
    4. Amorim P
    5. Janavs J
    6. Weiller E
    7. Hergueta T
    8. Baker R
    9. Dunbar GC

    (1998)

    The Mini-International neuropsychiatric interview (MINI): thedevelopment and validation of a structured diagnostic psychiatric interview forDSM-IV and ICD-10

    The Journal of Clinical Psychiatry 59 Suppl 20:22–23.

    • PubMed
    • Google Scholar
63. 1. Smith EE
    2. Reznik SJ
    3. Stewart JL
    4. Allen JJ

    (2017) Assessing and conceptualizing frontal EEG asymmetry: an updated primer on recording, processing, analyzing, and interpreting frontal alpha asymmetry

    International Journal of Psychophysiology 111:98–114.

    https://doi.org/10.1016/j.ijpsycho.2016.11.005

    • PubMed
    • Google Scholar
64. 1. Smith EE
    2. Cavanagh JF
    3. Allen JJB

    (2018) Intracranial source activity (eLORETA) related to scalp-level asymmetry scores and depression status

    Psychophysiology 55:e13019.

    https://doi.org/10.1111/psyp.13019

    • Google Scholar
65. 1. Steegen S
    2. Tuerlinckx F
    3. Gelman A
    4. Vanpaemel W

    (2016) Increasing transparency through a multiverse analysis

    Perspectives on Psychological Science 11:702–712.

    https://doi.org/10.1177/1745691616658637

    • PubMed
    • Google Scholar
66. 1. Stewart JL
    2. Bismark AW
    3. Towers DN
    4. Coan JA
    5. Allen JJ

    (2010) Resting frontal EEG asymmetry as an endophenotype for depression risk: sex-specific patterns of frontal brain asymmetry

    Journal of Abnormal Psychology 119:502–512.

    https://doi.org/10.1037/a0019196

    • PubMed
    • Google Scholar
67. 1. Stewart JL
    2. Coan JA
    3. Towers DN
    4. Allen JJ

    (2014) Resting and task-elicited prefrontal EEG alpha asymmetry in depression: support for the capability model

    Psychophysiology 51:446–455.

    https://doi.org/10.1111/psyp.12191

    • PubMed
    • Google Scholar
68. 1. Stordal E
    2. Mykletun A
    3. Dahl AA

    (2003) The association between age and depression in the general population: a multivariate examination

    Acta Psychiatrica Scandinavica 107:132–141.

    https://doi.org/10.1034/j.1600-0447.2003.02056.x

    • PubMed
    • Google Scholar
69. 1. Sun S
    2. Li X
    3. Zhu J
    4. Wang Y
    5. La R
    6. Zhang X
    7. Wei L
    8. Hu B

    (2019) Graph theory analysis of functional connectivity in major depression disorder with High-Density resting state EEG data

    IEEE Transactions on Neural Systems and Rehabilitation Engineering : A Publication of the IEEE Engineering in Medicine and Biology Society 27:429–439.

    https://doi.org/10.1109/TNSRE.2019.2894423

    • PubMed
    • Google Scholar
70. 1. Szumska I
    2. Gola M
    3. Rusanowska M
    4. Krajewska M
    5. Żygierewicz J
    6. Krejtz I
    7. Nezlek JB
    8. Holas P

    (2021) Mindfulness-based cognitive therapy reduces clinical symptoms, but do not change frontal alpha asymmetry in people with major depression disorder

    International Journal of Neuroscience 131:453–461.

    https://doi.org/10.1080/00207454.2020.1748621

    • Google Scholar
71. 1. Tadel F
    2. Baillet S
    3. Mosher JC
    4. Pantazis D
    5. Leahy RM

    (2011) Brainstorm: a user-friendly application for MEG/EEG analysis

    Computational Intelligence and Neuroscience 2011:1–13.

    https://doi.org/10.1155/2011/879716

    • PubMed
    • Google Scholar
72. 1. Thibodeau R
    2. Jorgensen RS
    3. Kim S

    (2006) Depression, anxiety, and resting frontal EEG asymmetry: a meta-analytic review

    Journal of Abnormal Psychology 115:715–729.

    https://doi.org/10.1037/0021-843X.115.4.715

    • PubMed
    • Google Scholar
73. 1. Tibshirani RJ
    2. Efron B

    (1993)

    An introduction to the bootstrap

    Monographs on Statistics and Applied Probability 57:1–436.

    • Google Scholar
74. 1. Toldo R
    2. Gherardi R
    3. Farenzena M
    4. Fusiello A

    (2015) Hierarchical structure-and-motion recovery from uncalibrated images

    Computer Vision and Image Understanding 140:127–143.

    https://doi.org/10.1016/j.cviu.2015.05.011

    • Google Scholar
75. 1. Tomé AM

    (2006) The generalized eigendecomposition approach to the blind source separation problem

    Digital Signal Processing 16:288–302.

    https://doi.org/10.1016/j.dsp.2005.06.002

    • Google Scholar
76. 1. Vallat R

    (2018) Pingouin: statistics in python

    Journal of Open Source Software 3:1026.

    https://doi.org/10.21105/joss.01026

    • Google Scholar
77. 1. van der Meij R
    2. Jacobs J
    3. Maris E

    (2015) Uncovering phase-coupled oscillatory networks in electrophysiological data

    Human Brain Mapping 36:2655–2680.

    https://doi.org/10.1002/hbm.22798

    • PubMed
    • Google Scholar
78. 1. van der Meij R
    2. van Ede F
    3. Maris E

    (2016) Rhythmic components in extracranial brain signals reveal multifaceted task modulation of overlapping neuronal activity

    PLOS ONE 11:e0154881.

    https://doi.org/10.1371/journal.pone.0154881

    • PubMed
    • Google Scholar
79. 1. van der Vinne N
    2. Vollebregt MA
    3. van Putten MJAM
    4. Arns M

    (2017) Frontal alpha asymmetry as a diagnostic marker in depression: fact or fiction? A meta-analysis

    NeuroImage: Clinical 16:79–87.

    https://doi.org/10.1016/j.nicl.2017.07.006

    • Google Scholar
80. 1. van Ede F
    2. Maris E

    (2016) Physiological plausibility can increase reproducibility in cognitive neuroscience

    Trends in Cognitive Sciences 20:567–569.

    https://doi.org/10.1016/j.tics.2016.05.006

    • PubMed
    • Google Scholar
81. 1. van Es MWJ
    2. Schoffelen JM

    (2019) Stimulus-induced gamma power predicts the amplitude of the subsequent visual evoked response

    NeuroImage 186:703–712.

    https://doi.org/10.1016/j.neuroimage.2018.11.029

    • PubMed
    • Google Scholar
82. 1. Van Veen BD
    2. van Drongelen W
    3. Yuchtman M
    4. Suzuki A

    (1997) Localization of brain electrical activity via linearly constrained minimum variance spatial filtering

    IEEE Transactions on Biomedical Engineering 44:867–880.

    https://doi.org/10.1109/10.623056

    • PubMed
    • Google Scholar
83. 1. van Vliet M
    2. Liljeström M
    3. Aro S
    4. Salmelin R
    5. Kujala J

    (2018) Analysis of functional connectivity and oscillatory power using DICS: from raw MEG data to Group-Level statistics in Python

    Frontiers in Neuroscience 12:586.

    https://doi.org/10.3389/fnins.2018.00586

    • PubMed
    • Google Scholar
84. 1. Vuga M
    2. Fox NA
    3. Cohn JF
    4. George CJ
    5. Levenstein RM
    6. Kovacs M

    (2006) Long-term stability of frontal electroencephalographic asymmetry in adults with a history of depression and controls

    International Journal of Psychophysiology 59:107–115.

    https://doi.org/10.1016/j.ijpsycho.2005.02.008

    • PubMed
    • Google Scholar

Author details

1. Aleksandra Kołodziej

   University of Social Sciences and Humanities, Warsaw, Poland

   Contribution

   Conceptualization, Resources, Data curation, Formal analysis, Funding acquisition, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing

   For correspondence

   akolodziej@swps.edu.pl

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-6042-8215

2. Mikołaj Magnuski

   University of Social Sciences and Humanities, Warsaw, Poland

   Contribution

   Conceptualization, Data curation, Software, Formal analysis, Visualization, Methodology, Writing - original draft, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-6859-2581

3. Anastasia Ruban

   University of Social Sciences and Humanities, Warsaw, Poland

   Contribution

   Software, Formal analysis, Investigation, Visualization, Writing - original draft

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-7039-148X

4. Aneta Brzezicka

   1. University of Social Sciences and Humanities, Warsaw, Poland
   2. Cedars-Sinai Medical Center Department of Neurosurgery, Los Angeles, United States

   Contribution

   Conceptualization, Resources, Supervision, Funding acquisition, Methodology, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-1950-4180

• 4,953

  views

• 392

  downloads

• 35

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.