Except by way of speculation and opinion, the systematic psychology of intelligence remains almost untouched. Observationally one can say little more about intelligence than that it is what the intelligence tests test. So defined we can then go much further and state a variety of psycho-social capacities that depend in a measurable degree upon intelligence as the tests test it, but all this work has brought us little farther toward an understanding of the nature of intelligence, its meaning in psychological and physiological terms, and hence its position in the body of psycho-physiological fact. The present paper seeks to make from this point of view an initial attack upon the problem of speed in intelligence (as the tests test it). We wish to point out, however, that the experiments presented, although illustrative of certain points and indicative of others, are not sufficiently extended to determine conclusively the relation of intelligence to speed. We think that they indicate what the ultimate conclusion is likely to be, and that they point the way to a program of investigation.

Among intelligence tests we have the speed tests and the

1 From the Harvard Psychological Laboratory. The junior author has conducted the experiment, tabulated the data, and made most of the computations; the senior author has collaborated with her in the interpretation of the results.

* For example, see the Symposium on Intelligence and its Measurement by fourteen authors, J. Educ. Psychol., 1921, 12, 123-147, 195-216, 271–275.

71

3

all-or-none tests. Because psychologists have sought to find and to measure an intelligence that is independent of the speed of reaction of the subject, the all-or-none tests are often taken to be the ideal. The line between the two types of test can not, however, be sharply drawn. "Are 'allow' and 'permit' the same or opposite?" is a speed test for an English-speaking adult; whereas "Does turpentine come from petroleum, ore, hides, or trees?" is an all-or-none test for the same person, for he either knows or does not know. Moreover, the test that depends upon the meanings of 'allow' and 'permit' becomes an all-or-none test for a child of eight or for a German adult; and in general speed tests for adults are likely to become allor-none tests for children, and all-or-none tests for children are likely to be speed tests for adults.1

From these remarks it appears that the all-or-none test is not a test of human performance that does not involve speed, but is simply a test that does not take account of speed. Such tests may be extremely valuable, especially as tests of information, and the neglect of speed is justified where the purpose is to test the availability of information in situations where time is of no importance. The all-or-none tests tend, however, to become tests of accomplished learning, and thus to be educational tests, in a broad manner of speaking, rather than intelligence tests, except in so far as intelligence is held to imply educability, and education, with a normal exposure

3 'Rate' tests and 'power' tests are the more usual phrases; cf. W. S. Monroe, Introduction to Theory of Educational Measurement, 1923, 63; M. R. Trabue, 'Measuring Results in Education,' 1924, 331-333, 336 f. Miss Margaret Kennedy of the Harvard Graduate School of Education, in an experimental study, as yet unpublished, substitutes the phrase 'all-or-none' for 'power.' The former phrase is to be much preferred, for 'power' is well established in scientific usage as rate of work, and its application to the all-or-none test is an exact reversal of its accepted meaning. Cf. the last section of this paper.

Here we have especially in mind Miss Kennedy's investigation to which we have just referred.

The difficulty of distinguishing between intelligence tests and educational tests, or between the educational and intelligent factors that enter into performance in an intelligence test, is well known. Cf., e.g., C. Burt, Mental and Scholastic Tests, 1921, esp. 175-183; Trabue, op. cit., 318-320; R. Pintner, Intelligence Testing, 1923, 60 f., 160-163; Psychological Examining in U. S. Army, Mem. Nat. Acad. Sci., 1921, 15, 764 ff., 779 ff.; etc. Professor W. F. Dearborn's Lowell Lectures (as yet unpublished) are also in point.

to social influences, is thus believed to measure intelligence. Even here, however, it may be pointed out that something might be gained were the all-or-none tests not to neglect time. If these tests measure learning, so does the method of right associates, and Müller and Pilzecker found that this method was greatly sensitized by adding to it a measurement of the association reaction times.6

The essence of the intelligent reaction seems to be the solution of a problem against time. The most obvious and the most general variable in this reaction is the time, although it is plainly not the only important variable. If the organism may not be prepared to solve the problem at all in any amount of time, then we have the all-or-none test. If, however, a subject solves an all-or-none test, then clearly the time becomes important: it is not so good a solution for him to solve it slowly as to solve it rapidly; and neglect of differences in time is thus a neglect of important differentiæ. In fact the all-or-none test may be regarded simply as the limit of the fast-or-slow test, the limit where the time required is recorded simply as finite or infinite. It is plain that the all-or-none test, measuring an essentially temporal act, nevertheless sets conditions which force it to disregard times because the times obtained are of the least precise order.

At any rate, speed is so important in the intelligent act that it has seemed to us the first factor to be studied if we are ultimately to come at an understanding of the nature of intelligence."

In respect of speed there are various problems. If the speed of the intelligent act can be isolated, there arises the problem of its identification with intelligence as the tests test it. Will pure speed tests correlate as highly with psychological and social capacities as have intelligence tests? If they will not, what other factors must be considered?

The Treffer- und Zeitmethode of G. E. Müller and A. Pilzecker, Experimentelle Beiträge zur Lehre vom Gedächtniss, 1900. The Treffermethode by itself is practically an all-or-none mental test, yet Müller and Pilzecker saw the importance of taking account of time.

7 E. L. Thorndike proposes the analysis of ability into level, range, and speed: J. Educ. Research, 1925, 11, 1-11, esp. 9 f. Against his tendency to view these three as at least equally important, we set the argument of the text for the fundamental relation of speed to mental activity.

There is also, however, the problem of the locus of the speed, a problem which points more directly toward the fundamental nature of intelligence. Suppose we find that the man who does poorly in an intelligence test can do just as well as the man who did well, provided he is given more time than the other man. If there are no items of the all-or-none kind, he may, of course, if given more time, do just as well. Now the problem of intelligence, as such a test tests it, has become entirely a matter of speed; and the first thing to do, we think, is to localize the wasted time.

There are two possibilities. The loss of time by the 'less intelligent' subject may be interstitial or it may be inherent in the intelligent act. It would be interstitial if the good and the poor subject performed every required operation at equal speed, but the poor subject was less able to limit himself to the essential requirements and lost time by irrelevant activities or by self-distraction. Everyone knows how conscious effort itself may become a distraction from the task upon which the effort is directed, and it is possible that a great deal of irrelevant 'thought' accounts for the slowness of the subject who does poorly in an intelligence test. On the other hand, the loss of time may be said to be inherent in the intelligent act, if, as far as analysis can be pressed, it is found that the constituents of the act occur more slowly in the poor subject than in the good subject. It is with this problem that we are concerned in the present paper.

The first step in analysis consists in measuring the times of the separate items of an intelligence test. We must find out whether in the Army Alpha Examination, for example, there is a tendency for the poor subject to do every one of the 212 items more slowly than does the good subject. If he does in general, then it is clear that any distractions that cause him to lose time are not of the gross order that could be revealed by introspection under ordinary introspective conditions. If, however, the poor subject equals the good subject in some stretches of the test, but falls behind on account of abrupt retardations, then an appeal might readily be made to an introspective investigation.

If, item for item, the poor subject tends to be slower than the good subject, then we must push our analysis within the item. Our first interest will be in the simplest items, like those of the 'same-and-opposite' test. If the single items of this test show the same time-difference as does the examination as a whole, we can see whether we shall find the same individual differences in associative reaction times, in cognitive reaction times, in simple reaction times, in reflex times, and ultimately, if necessary, in the rates of nerve-conduction. If it be argued that we should not expect a difference in the rates of nerve conduction, then we may point out that somewhere in the series of times just listed the time-differences, apparent in the intelligence test, would disappear, and we should have localized the level of complexity at which the difference in time appears. Such a localization we think of as a first step toward the solution of the problem of the nature of intelligence.

THE EXPERIMENT

We had five laboratory subjects: A, B, C, D, and E. Two of the subjects were men, graduate students of psychology; the other three were women, advanced undergraduates, conducting experimental problems in the Harvard Psychological Laboratory. Laboratory subjects of this sort are presumably a highly selected group in respect of intelligence. Laboratory conditions, however, are ideal for precise measurement, and the consequent gain in precision may render significant the small difference in intelligence that the tests reveal in the selected group. We think that the present paper demonstrates the feasibility of studying intelligence in the laboratory with a few selected subjects, even though it would be necessary ultimately to check the conclusions formed in the laboratory by recourse to a larger and more heterogeneous group.

These subjects were all given forms 5 and 6 of the Army Alpha Examination, and forms A and B of the Higher Examination of the Otis Self-Administering Tests of Mental Ability.

These letters were assigned to the subjects after the computation of the data so that the order A, B, C, D, E might be also the average order of intelligence and of speed; cf. Table VI.