Dear Nastassja,

It depends whether the EEG frequencies relate to the same subject or different subjects. For the former use repeated measures, for the latter use factorial.

Enhancing Peter's answer, if you have a within subjects factor (like repeated measurements from the same subject) then you should perform a version of RM-ANOVA. If you have a between subjects factor (like different groups) then you should perform an ANOVA (may be factorial). If you have both, that ANOVA is called mixed.

Apparently, you have a two-way factorial design. This can be a 'between x between', 'between x within' (also named 'mixed'), or 'within x within' (doubly repeated) factorial design. The term between  refers to a between-subjects independent factor, for which a different (randomly assigned) group of subjects (or units of observation, or participants) is used for each level of the factor. A within-subjects factor, on the other hand, is an independent factor that is manipulated by testing each subject at each level of the factor, also named repeated measures. A RM-ANOVA usually is a 'mixed' ('between x within') design ANOVA.

Any chance for mixed effect model here

Mixed effect models is a superset of ANOVA.

Could you give a more detailed description of the study. What did you measure (dependent variables/responses, many times on the same subjects? = repeated measures), what did you manipulate (factors/ independent variables)?

A repeated measures model is only relevant if there is a correlation between the measure at time A and the measure at time B. So, I measure EEG for 1 hour on person 1 and then do that again after a 10 minute break. I would expect a high degree of correlation. On the other hand, if I wait 23 years, so much has happened that there shouldn't be any correlation. The rate at which correlation breaks down over time is unknown (to me). Could you still detect a correlation after 1 week, 1 month, 1 year?

You could run a repeated measures design and then run standard ANOVA if the repeated measures factor is not significant. However, this is not good statistical practice because what you are saying is that a failure to find a significant difference is proof of the null hypothesis. However, in this case you know that there will come a time when the null hypothesis will be true. It will be preceded by a time when the null-hypothesis is false, but it will take 200 billion samples to achieve significance (the null-hypothesis is effectively true). The correlation will gradually increase with shorter and shorter time intervals until the correlation is detectable even with your sample size.

It is better to use a repeated measures ANOVA when the measures are from the same persons at different times, even when the correlation is not significant.

See, e.g.:

http://blog.minitab.com/blog/adventures-in-statistics-2/repeated-measures-designs-benefits-challenges-and-an-anova-example

"The Benefits of Repeated Measures Designs

More statistical power: Repeated measures designs can be very powerful because they control for factors that cause variability between subjects."

Factorial and repeated-measures ANOVA are not in opposition. A factorial ANOVA can be repeated-measures or not; a repeated-measures ANOVA can be factorial or not.

Dear Nastassja,

If you are working with repeated measures of a dependent variable, in intragroup replication, i.e., two related samples (each subjects as your own control) or paired groups, and you want to compare the effect of two factors on that dependent variable, you can use "ANOVA two-factor experiment with repeated measures on one factor" (see: Winer BJ, Brown DR, Michels KM. (Eds.) Statistical Principles in Experimental Design, 3rd ed. 1991, McGraw-Hill: p.509 - 531), as known as "ANOVA for profile data" (see: Morrison DF. (Ed.) Multivariate Statistical Methods, 3rd ed. Mc-Graw-Hill, 1990: p. 243-254). Sincerely,

Otavio Pimentel Cesar