Appropriate error bars for repeated-measurements designs I conducted a t-test / ANOVA (both repeated measurements) and I want to represent the difference in the mean via a bar graph.
There are several different views about the appropriate error bars for repeated-measurements: personal preference (Field, 2009), Root Mean Square Error (Estes, 1997) or rather Statistical Significance Bars (Schunn, 1999).
What is the best solution?
 A: A few quick thoughts:


*

*The most important thing is that you state clearly what the error bar represents. Too often, it is left up to the reader to guess what the error bars might represent (e.g., confidence intervals or standard errors? between subjects or within subjects standard errors?).

*I think Estes' paper provides some good advice.

A: I must admit I'm unfamiliar with Field 2000 but I concur with Jeromy Anglim about Estes and that you must clearly label.  
My recommendation is that you plot the effect and within S confidence interval around the effect.  In your text include a standard error for the overall mean for meta-analysis purposes but downplay it.  
Conveying between S calculated error bars of any kind with repeated measures is generally unwise because that's not what you tried to measure and often the estimates of the means across subjects are very variable with repeated measures because the N is low (you may have a high N but generally repeated measures experiments have low Ns).
