Subsets of vectors or data frames

Note that:

na.omit(x) generates x, less NAs.  But it
does more.  It keeps traceback information
on what elements were omitted.

> x <- c(1, 5, NA, 2, 0, 11)
> xy <- data.frame(x=x, y=letters[1:6])
> na.omit(x)
[1]  1  5  2  0 11
attr(,"na.action")
[1] 3
attr(,"class")
[1] "omit"

> ## Try also na.omit(xy)

The object that is returned retains traceback 
information, in the attributes, on what has 
been done.

na.exclude() has almost the same effect [the
difference affects the way that information
in the object may be returned later]
> na.exclude(x)
[1]  1  5  2  0 11
attr(,"na.action")
[1] 3
attr(,"class")
[1] "exclude"

Information on the class is used by some 
modeling functions to determine whether residuals 
etc are returned for all data [class="exclude"]
or [with class="omit"] only for non-missing data.

Or you can do, e.g.

> keep <- !is.na(x) & x!=0
> x[keep]
[1]  1  5  2 11

> ## Alternative
> subset(x, keep)
[1]  1  5  2 11

> ## Try also
> # xy[keep, ]
> # subset(xy, keep)


Part V, 4(d)
To regress log(weight) on log(thick) and 0.5*(log(height)+log(breadth)),
the model formula needed is
log(weight) ~ log(thick) + I(0.5*(log(height)+log(breadth)))

The reason for the use of the wrapper function I() is to prevent the
parser from giving "*" the special meaning that it has in a model formula.


Part VI

Note that fitted() and residuals() and other such functions
are applied to model objects, i.e., for lm models, to the
output from lm(....)