You've understood me perfectly. I want the interaction and there will be 6 values graphed. Now I understand why the multiplication by 2 (which is unnecessary).
To restate what you said, if I want an error bar (1 SE) for the interaction it's simply: Sqrt(MSEinteraction/18) Correct? There will be six bars and each bar represents 18 subjects. Allyson -----Original Message----- From: Donald Burrill [mailto:dfb@;mv.mv.com] Sent: Friday, October 25, 2002 3:56 PM To: Allyson Cc: [EMAIL PROTECTED] Subject: Re: error bars on histogram for ANOVA interaction Allyson, there are a couple of points that may need to be clarified: especially if I have incorrectly interpreted what you meant (and I'll try to be clear about what I've inferred between the lines). On 25 Oct 2002, Allyson wrote: > I'm trying to plot error bars from the results of an ANOVA (Group[2x18 > people in each group], Effect[3 levels repeated measure]). I'd like > to use the MSError to do this. Please tell me if I'm correct First clarification needed: whether the means you want to plot (or so one presumes -- those are what error bars are commonly attached to) are the means of the Group main effect, of the "Effect" main effect, or of the Group-X-Effect interaction. Your "Subject:" line mentions "interaction", so I'll assume that's what you meant: that is, that you want to display a graph of six different means (whose values are plotted on the ordinate of the graph): a thing that might look something like this: + - G1 - G1 - - + G2 - G1 - - - G2 + G2 - --------+---------+---------+---- E1 E2 E3 with (vertical) error bars on the plotted points (that I've labelled G1 and G2, for the pertinent Group); and you might want to connect the three "G1" points with straight lines, and similarly for G2, which is not something I can do in this medium. > to generate an error bar from the MSError from an effect: > sqrt((MSE*2)/n) You haven't said what kind of error bar you have in mind. Sometimes one wants (mean +/- 1 standard error of the mean), which would not entail your multiplication by 2; sometimes one wants to represent a 95% (or in general a [1-alpha] confidence interval, which would entail multiplying the standard error by, approximately, 2 -- but after one had taken the square root, not before. It is not clear to me why MSE would be doubled. The value for "n" is the number of cases represented by each mean to be displayed: for the G effect that would be 18*3 = 54; for the E effect it would be 18*2 = 36; for the GE (interaction) effect it would be 18. (Hence your "36*3" in the next line is incorrect, regardless of whether you meant it to be enclosed in parentheses (so as to divide (2*MSE) by 108) of not (so as to divide (2*MSE) by 12, effectively).) > errorBarValueOfInteraction=sqrt((MSEOfIntraction*2)/36*3) > > Then each bar value in the histogram of component scores would have > the same sized error bar? Yes. You would be using the pooled variance estimate for the E and EG effects. (Note: the MSE for G is not the same as the MSE for E and EG in this repeated-measures design. The former is likely to be an order of magnitude larger than the latter, at least in the usual situation where Ss that score high on one of the repeated measures tend also to score high on the others.) > Forgive me if I'm clueless. We all were, once; and still are, some of the time, about some things at least. I hope this has been helpful. If I've incorrectly understood what you intended, please let me know; if you don't understand something I've written, ask for clarification. (I'm not TRYing to be obscure!) -- DFB. ----------------------------------------------------------------------- Donald F. Burrill [EMAIL PROTECTED] 56 Sebbins Pond Drive, Bedford, NH 03110 (603) 626-0816 [was: 184 Nashua Road, Bedford, NH 03110 (603) 471-7128] . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
