Question on ezANOVA within_full

[absurd]

Hi,

I just realized the (relatively) new option within_full in ezANOVA. I ran an Rscript that I previously used on some data and got a lot of these warnings:

Collapsing data to cell means. *IF* the requested effects are a subset of the full design, you must use the "within_full" argument, else results may be inaccurate.

Well, I'm thinking about what this actually means and if this might be crucial for me. Could somebody give me an example that makes clear in which case using within_full instead of within matters? What is really going on "behind the scenes" in this case?

Thanks for your help.

J.

[Mike Lawrence]

Wow, I really dropped the ball in responding promptly on this one! Sorry about that :O\

The within-full argument is there so that, when dealing with data that needs to be collapsed to cell means, you get the same results no matter what within-Ss variables you enter. For example, say you have a design with two within-Ss variables, A & B, but the data are unbalanced at the level of the raw data (ex. you had to toss observations from specific conditions from specific subjects). The full anova would be:

    ezANOVA(

        data = my_data

        , dv = my_dv

        , wid = subject

        , within = .(A,B)

    )

But if you ran an anova with just one of the variables and failed to pass it to within full, as in:

    ezANOVA(

        data = my_data

        , dv = my_dv

        , wid = subject

        , within = A

    )

You might get different results for the main effect of A as compared to the previous call. This is because the second call collapses the data to a mean for each level of A within each subject, and if the number of observations vary between levels of B, this affects the means. So, the proper call is:

    ezANOVA(

        data = my_data

        , dv = my_dv

        , wid = subject

        , within = A

        , within_full = B

    )

The use of this really isn't for doing anovas, as you'd more typically simply do the full anova and leave it at that. More useful is plotting main effects from a complicated design, where ezPlot gets ezANOVA to collapse the data and provide means and error bars. So, if I wanted to look at the plot of the main effect of A, I'd do:

    ezPlot(

        data = my_data

        , dv = my_dv

        , wid = subject

        , within = A

        , within_full = B

        , x = A

    )