文档介绍：Chapter 15—Hypothesis-Testing with Counted Data, Part 1 205
CHAPTER
Hypothesis-Testing
with Counted Data,
15 Part 1
Introduction
Should a Single Sample of Counted Data be Considered Different From a
Benchmark Universe?
Introduction
The first task in inferential statistics is to make one or more
point estimates—that is, to make one or more statements about
how much there is of something we are interested in—includ-
ing especially the mean and the dispersion. (That work goes
under the label “estimation” and is discussed in chapter 13.)
Frequently the next step, after making such quantitative esti-
mation of the universe from which a sample has been drawn,
is to consider whether two or more samples are different from
each other, or whether the single sample is different from a
specified value; this work goes under the label “hypothesis
testing.” We ask: Did something happen? Or: Is there a differ-
ence between two universes? These are yes-no questions.
In other cases, the next step is to inquire into the reliability of
the estimates; this goes under the label “confidence intervals.”
(Some writers include assessing reliability under the rubric of
estimation, but I judge it better not to do so).
So: Having reviewed how to convert hypothesis-testing prob-
lems into statistically testable questions in Chapter 14, we now
must ask: How does one employ resampling methods to make
the statistical test? As is always the case when using resampling
techniques, there is no unique series of steps by which to pro-
ceed. The crucial criterion in assessing the model is whether it
accurately simulates the actual event. With hypothesis-testing
problems, any number of models may be correct. Generally
speaking, though, the model that makes fullest use of the quan-
titative information available from the data is the best model.
When attempting to deduce the characteristics of a universe
from sample data, or when asking wheth