Study Statistics and Visualization Options
Advanced statistics are performed in LIMO (Linear Modeling of EEG data), an EEGLAB plugin, primarily developped by Cyril Pernet in collaboration with Arnaud Delorme. The LIMO toolbox allows you to use general linear modeling approaches on an arbitrarilly large number of categorical and continuous variables. The EEGLAB team have recently developed a more user friendly interface for LIMO, that directly interfaces EEGLAB variables. The documentation about the old version of LIMO is available here. You may also refer to the LIMO tutorial video series.
Difference between optimization methods
 Ordinary Least Square (WLS): This is the simplest and fastest method. Same as using the MATLAB glmfit function.
 Weighted Least Square (WLS): The default. This method attributes some weight to individual trials based their outlier likelyhood.
 Iterated Reweighted Least Square (IRLS): The default. This method attributes some weight to individual trials based their outlier likelyhood.
LIMO Faq.
These questions were answered by Cyril Pernet.

Question: Are the bootstrap only 2nd level or can you do it 1st level as well when you look at a single subject? Bootstrap statistics are only at the second level.

Question: Are contrast only for 2nd level posthoc on ANOVA analyses? No, contrast can be calculated at the first level as well.

Question: Is there a difference between a paired ttest at the group level between condition 1 and 2, and computing a contrast between condition 1 and 2 at the single subject level, and then computing a onesample ttest at the group level on that difference? In theory this is equavalent. However, LIMO uses robust statistics, and uses Yuen ttest at the group level. It is preferable to use the ttest at the group level.

Question: Can I process more than 2 groups of subjects in LIMO? You can, but you cannot do more than 2 nested groups at the time for secondlevel statistics.

Question: What is the limit in terms of number of 1stlevel variables? There is no limits. You can have as many categorical and continuous variables as you want.

Question: What do I sometimes get errors when using Weighted Least Square (WLS) optimization at the 1st level? WLS requires that you have more trials than samples. You can reduce the time window or frequency range to decrease the number of samples. This is not a problem with Ordinary Least Square.