080 PONTIFICIAE ACADEMIAE SCIENTIARVM SCRIPTA VARIA - 28

test performed by Arrais. I can well understand that one cannot
rest much on an argument such as this one, since these numbers
are very rough and are subject to substantial errors of measurement.
 Indeed, as ALLAIS has shown, even plausible errors of
measurement can lead to this result. On the other hand, I am
anable to understand why these numbers are appropriate for ALLAIS
to use and not appropriate for me to criticize. Professor ALLAIS
.nsists upon saying that everything goes according to his model.
In fact, this does not go according to his model, and while that
may be the result of measurement error, it may equally not be. It
s, of course, all too common for an investigator to claim that numders
 which apparently point opposite to his theory are subject to
measurement error. Professor ArLals, however, is managing to do
more than that. First he puts forth these numbers as supporting his
theory; then, when I point out that in fact they do not support it,
he claims that is due to measurement error; finally, he claims that
n fact they do support it. Professor ALLAIS is working both sides
of the street here and I am afraid I cannot understand his somewhat
engthy argument on this point.

ALLAIS

First, in my opinion it is impossible to attach any importance
to these slight differences because the precision of the different
estimates is quite small and it is difficult to derive any conclusion
from small deviations which are probably of a random character.
For instance, if you consider the 58 Corin CLARK figures for
21 countries cited by me in Table V, p. 54 of my Tokyo paper
(ALLAIS, 1960 A), the range for y is quite large, from 0.83 to 8.48.
The median is 3.54 and the coefficient of variation about it is 22%.
If you consider each of the 58 figures, this is certainly wrong
and only the median of the lognormal distribution has any real
meaning.

11] Allais - pag. 284