Greetings.
Here is a problem I don't know how to handle, even by brute force. We have
an 800k line data file that includes eye fixations for subjects in a 3 x 2
factorial design. There are several screen locations where information is
available while Ss do their task. These locations vary by condition so there is
no reason for people in some conditions (i.e., the 3-factor one) to look at some
locations. So I am analyzing these conditions, two pairs at a time (i.e., the
2-factor one).
So the problem: My ANOVA of these data had messed up the within and between
factors the way aov does when there are different numbers of Ss contributing
data to some of the variables than for others. A bit of sleuthing with plyr
revealed that one of our Ss in one of our conditions never looked at one of our
locations.
Given the nature of the DV, zero is a fine number. Although a little unexpected
in this condition, it is reasonable and cannot be ignored. However, R
recognizes, rightly, that lack of data is not the same as zero.
I guess I could "subset" and "aggregate" the dataset to pull
out a data.frame that contains the data aggregated at the right level for this
aov and then add one record for this one Ss (well - actually I would add 8
records as "block" is a within-Ss factor that we have been looking
at). But there must be a more elegant way of doing this especially as we are
still in the exploratory phase and will be pulling out factors such as mean
dwell time, total dwell time (dwell time per fix, summed over all fix) and other
factors (e.g., tallies of sequential fixation chains -- e.g., obj-A, obj-B, vs
obj-A, obj-C, etc, etc, etc).
As always, your thoughts, comments, and code would be appreciated.
Thanks,
Wayne Gray
