r. Schillings and myself. The
numbers 2, 3 and 4, on the other hand, were very easily obtained and,
above all, 3 seldom failed. 3 seemed to be the horse's favorite number
and was very frequently given instead of other numbers. Thus, one-sixth
of all the horse's incorrect responses which were given to me were in
terms of the number 3. The numbers 5 and 6 were a little more difficult
to obtain and above 10 the difficulty increased rapidly. Indeed, I never
saw Hans respond with a number exceeding 20 to any questioner, Mr.
Schillings and Mr. von Osten excepted. I saw the nine vain attempts of
Count zu Castell to get the number 15, and Count Matuschka's eight
unsuccessful attempts to obtain the number 16 as a response. But even
with Mr. von Osten and Mr. Schillings such failures were not infrequent.
Thus, Mr. von Osten tried five consecutive times to obtain the number
24. I myself did not fare any better at first. But the following table
shows what practice can do. If we compare the percentage of correct
responses (involving the numbers 1 to 7--for which alone I have
sufficient material, viz., 80 to 100 cases), obtained in the first half
of our tests, with that of the second half, we get the following:

For number: 1 2 3 4 5 6 7
In first half of tests: 49, 92, 89, 86, 74, 62, 53%
" second " " " : 92, 95, 92, 98, 97, 86, 96%

From this we see how hard it was at first to get the number 1 and that
failure was as frequent as success, and how much easier it was on the
other hand to get the numbers 2 and 3 (and which, therefore, do not show
any great improvement in the second half of the tests). Beyond the 3 the
percentage of correct responses decreased and the number 7 stood at the
same level as the number 1. In the second half of the tests, all these
differences disappeared and errors were infrequent and seldom exceeded
+1 or -1. These results of practice are not to be accredited to the
horse, but to the experimenter, who was at first quite unskilled. This
difference in results does not appear in the case of Mr. von Osten, for
his initial practice had been had many years previous. The values
obtained in his case were very constant throughout our experimentation
and generally showed something like 90% of correct responses. To be
sure, in his case also, the number 1 was somewhat unfavorable, (79% were
correct responses). But the percentages obtained in his case showed no
impr