Hi Alex!

Based on the structure of your dataset and the goals of your analysis, you could use a multivariate analysis of variance (MANOVA). It's appropriate when you have multiple dependent variables (in your case, the values linked to specific channel, freqband, and measure combinations) and you want to test for differences between two or more groups.

To identify significant differences at the channel, measure, or freqband level, you could consider conducting post-hoc tests or follow-up univariate analyses (e.g., ANOVA) for each dependent variable (channel, measure, or freqband) after significant results are found in the MANOVA.

Please ensure that the assumptions of MANOVA are met, such as multivariate normality, homogeneity of variance, absence of multicollinearity, and independent observations.

Before choosing MANOVA, take a look at the following.

http://psychologicalstatistics.blogspot.com/2021/08/i-will-not-ever-never-run-manova.html

Article MANOVA: A Procedure Whose Time Has Passed?

Article Multivariate Analysis Versus Multiple Univariate Analyses

Concur with the issues of using the MANOVA.

Assuming your data is linear, I recommend multiple regression modeling. You can check for cause and effect by separating out your variables as independent (cause) and dependent (effect). In other words independent variables are those you (as the researcher) are manipulating and dependent variables are those you are measuring a a result of that manipulation. You need to research the best way to add your variables into various models to find the greatest cumulative change of the dependent variables (R-squared and significance - sig.). By going through multiple models, you will find which dependent variables are effected by the independent variables. Also recommend looking at residual vs. fitted plots (typical X-Y charts) to confirm linearity.

Hope this is a helpful in narrowing down your statistical test(s). Make sure you keep track of EVERYTHING you do, to capture in the Methodology section of what ever research manuscript you may create (article, thesis, dissertation, etc.)

Hi Alex,

It is not uncommon when working with real-world data to be confronted with violations of one or more of MANOVA assumptions. Assumptions that should hold true for results of your MANOVA test to be trusted include:

OUTLIERS. Outliers are not inherently bad. Discarding them is usually a judgment call. Often times, the researcher may choose to overlook true outliers that represent natural variations in the population. Remove outliers only when you have a sound reason for doing so, particularly, those that are problematic - measurement errors, data entry errors, processing errors, or poor sampling. In SPSS, you can use boxplot, a good graphical visualization to detect univariate outliers. Similarly, you can check for multivariate outliers by using Mahalanobis distance.

NORMALITY. Unfortunately, multivariate normality cannot be directly tested in SPSS Statistics. Instead, you can use each of the dependent variables for each of the groups of the independent variables to confirm whether there is multivariate normality. For a numerical means of assessing normality where charts wouldn’t be enough, Shapiro–Wilk test is more appropriate method for small sample sizes (n

For your scenario, a suitable statistical test would be a repeated measures ANOVA or a linear mixed-effects model, considering the nested structure of your data (channels within subjects). This test can assess differences in values between groups while accounting for the repeated measures within subjects. Additionally, you can perform post-hoc tests to identify significant differences at the channel, measure, or freqband level.