[R] Help with Reshaping from Wide to Long

[Kingsford Jones]

On Sun, Jul 18, 2010 at 6:44 PM, jlwoodard <john.woodard at wayne.edu> wrote:
>
> Hi Phil and Jeff,
>    Thanks so much for taking the time to help me solve this issue!  Both
> approaches work perfectly.  Each of your approaches helped me learn more
> about what R can do.   I really appreciate your help!

Hi John,

Now that you've seen some of R's fancy data manipulation footwork,
here's a small taste of the graphing capabilties (the matching of
colors and glyphs with your factor levels is serendipitous :-))

dat <- structure(list(Accuracy = c(95L, 80L, 100L, 90L, 100L, 100L,
95L, 85L, 100L, 90L, 100L, 100L, 60L, 55L, 80L, 45L, 75L, 45L
), Subject = structure(c(1L, 2L, 1L, 2L, 1L, 2L, 1L, 2L, 1L,
2L, 1L, 2L, 1L, 2L, 1L, 2L, 1L, 2L), .Label = c("subj101", "subj102"
), class = "factor"), Shape = structure(c(1L, 1L, 1L, 1L, 1L,
1L, 3L, 3L, 3L, 3L, 3L, 3L, 2L, 2L, 2L, 2L, 2L, 2L), .Label = c("Circle",
"Triangle", "Square"), class = "factor"), Color = structure(c(3L,
3L, 1L, 1L, 2L, 2L, 3L, 3L, 1L, 1L, 2L, 2L, 3L, 3L, 1L, 1L, 2L,
2L), .Label = c("Red", "Green", "Blue"), class = "factor")), .Names =
c("Accuracy",
"Subject", "Shape", "Color"), row.names = c(NA, -18L), class = "data.frame")

library(ggplot2)
X11(12, 6)
qplot(Shape:Color, Accuracy, data = dat, colour = Color, shape = Shape,
   facets = . ~ Subject)

Suggesting (keeping in mind here we have a sample of just 2 subjects):
i) lower accuracy for triangles
ii) lower accuracy for blue (subj102 triangls is an exception)
iii) the upper bound on accuracy is often reached.
iv) the upper bound may mask effects.  For example look at the color
effects for circles and squares -- for subj101 the green effect might
be masked by the upper bound.
v) there is Color:Shape interaction (e.g. the color effects differ for
triangles)
vi) there is likely between-subject variation in the mean and possibly
in effects as well.

As for analyses, my preference  for repeated measures is to use
likelihood-based rather than sums-of-squares based methods.  Usually
I'd recommend lme4::lmer OR nlme:lme starting with random Subject
intercepts (appears to really just be a RCBD, so a start might be
lmer(Accuracy ~ Shape*Color + (1|Subject), dat)), but the constrained
response and limited sample size (both terms of number of subjects and
conflation between error and interaction) makes me think fitting a
meaningful model is not trivial.  Off the cuff, perhaps a beta or
binomial model or using logit-transformed Accuracy (noting that
nothing can retrieve the 'theoretical effects' mentioned in (iv)
above, but that may not be of interest),

best,
Kingsford Jones






>
> Very best regards,
>
> John
> --
> View this message in context: http://r.789695.n4.nabble.com/Help-with-Reshaping-from-Wide-to-Long-tp2292462p2293463.html
> Sent from the R help mailing list archive at Nabble.com.
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>