文档介绍：270 Resampling: The New Statistics
CHAPTER
The Statistics of
Hypothesis-Testing
18 With Measured Data
Chapters 15 and 17 discussed testing a hypothesis with data
that either arrive in dichotomized (yes-no) form, e as
data in situations where it is convenient to dichotomize. We
next consider hypothesis testing using measured data. Con-
ventional statistical practice employs such devices as the
“t-test” and “analysis of variance.” In contrast to -
plex devices, the resampling method does not differ greatly
from what has been discussed in previous chapters.
Example 18-1: The Pig Rations Still Once Again, Using
Measured Data (Testing for the Difference Between Means of
Two Equal-Sized Samples of Measured-Data Observations)
(Program “Pigs3”)
Let us now treat the pig-food problem without converting the
quantitative data into qualitative data, because a conversion
always loses information.
The term “lose information” can be understood intuitively.
Consider two sets of three sacks of corn. Set A includes sacks
containing, respectively, one pound, two pounds, and three
pounds. Set B includes sacks of one pound, two pounds, and
a hundred pounds. If we rank the sacks by weight, the two
sets can no longer be distinguished. The one-pound and
two-pound sacks have ranks one and two in both cases, and
their relative places in their sets are the same. But if we know
not only that the one-pound sack is the smallest of its set and
the three-pound or hundred-pound sack is the largest, but also
that the largest sack is three pounds (or a hundred pounds),
we have more information about a set than if we only know
the ranks of its sacks.
Rank data are also known as “ordinal” data, whereas data
measured in (say) pounds are known as “cardinal” data. Even
though converting from cardinal (measured) to ordinal
(ranked) data loses information, the conversion may increase
convenience, and may therefore be worth doing in som