で，
まっとうな仕事のほうは本日も進捗せず
1630 撤退．
JR 札幌駅周辺の古本屋・本屋をひさびさにハシゴ．
1920 帰宅．
[今日の素読]
Salsburg, D. 2001.
``
The Lady Tasting Tea
-- How statistics revolutionized science
in the twentieth century''.
Owl Book.
Chapter 16. Doing Away With Parameters
Unsolved problems
The development of nonparametric procedures may have led
to a burst of activity in this new field. However, there
was no obvious link between the parametric methods that
were used before this and the nonparametric mdethods.
There were two unsolved questions:
If the data have a know parametric
distribution like the normal distribution,
how badly will the analysis go wrong if we
use nonparametric methods?
If the data do not quite fit a parametric
model, how far off from that model must
the data be before the nonparametric
methods are the better ones to use?
By the time he began to work on his 1948 paper, Pitman had
developed a clear line of resoning about the nature of
statistical hypothesis tests and the interrelationships
between the older (parametric) and the newer
(nonparametric) tests. With his new methods, he attacked
the two outstanding problems.
What he found surprised everyone. Even when the original
assumptions are true, the nonparametric tests were almost
good as the parametric tests. Pitman was able to answer
the first questions: How badly do we do if we use
nonparametric tests in a situation where we know the
parametric model and should be using a specific parametric
test? Not badly at all, said Pitman.
The answer to the second question was even more surprising.
If the data do not fit the parametric model, how far off
from the parametric must they be for the nonparametric
tests to be better? Pitman's calculation'showed that with
only slight deviations from the parametric model, the
nonparametric tests were vastly better than the parametric
ones.
