yy ts nie 9) Bg Aue bad ‘¥ as Siiec aera! Ges oa “4 Fae Pag he peti i © ae te gwide i i a 4 Pstholeite iit: . i Sie), , inn i nat ts We if} ¥ AS TSEE. = us Dea heu i ant epya a Hiepae vudeeiy Pure ree ree to rie red tat} ie ; ee ¥ 9 ‘ ie a iseaar tea sy) ee fat fs i ; ‘f ae th 4 ate ar tit ‘ Hehe Hs ae ithe yh 7 : rod “4 : rade i Dat 4 =a ot pees Bt at PE AEE ze ae TeSasars: cree] yet eS 2 ene Sayeed eae rare x StS ere ee = PB ee apse at ceati CEO a * == ce BY? i) ot an rye * ee Me nite ‘ * a i eS ee = - wesc eeee! ante te PALL at ati ie ay is 4 AsO Uy treet’ Rites Wgiyk i ieee, f eaaety at iy | pL aha PARAS aA why i Ai Hints ee teeta ‘ Pr Pek, pect atEsees <2 uy re Creice ot ry t - 4 ye} +3 saat pl iy eies BL iy u L pe hs ahaa Rate ay isle Bae bank penstites ts Det ap Reber Digitized by the Internet Archive In 2022 with funding from Princeton Theological Seminary Library https://archive.org/details/relationofrateof0Ohigh /, y ear PSYCHOLOGICAL REVIEW PUBLICATIONS Fh herceep te 156 Psychological Monographs EDITED BY SHEPHERD I. FRANZ, Univ. or Cattr., So. Br. HOWARD C. WARREN, Princeton University (Review) JOHN B. WATSON, New York, N. Y. (J. of Exp. Psych.) MADISON BENTLEY, University or Ittinois (Index) and S. W. FERNBERGER, UNIversity or PENNSYLVANIA (Bulletin) SLIDES INePSYCHOLOGY FROM THE JESUP PSYCHOLOGICAL LABORATORY Relation of the Rate of Response to Intelligence we / BY y Pee UGH SMELT: Pre D. Professor of Psychology North Carolina College for Women PoverOLOGICAL’ REVIEW \COMPANY PRINCETON, N. J. AND ALBANY, N. Y. Acenis: G. E. STECHERT & CO., Lonpon (2 Star Yard, Carey St., W. C.); Lerpzic (Hospital St., 10); Parts (76, rue de Rennes) M i me z ACKNOWLEDGMENTS The writer wishes to acknowledge his indebtedness to those whose aid has made this study possible. He is especially indebted to Dr. Joseph Peterson of George Peabody College for Teachers for his encouragement and criticisms during the progress of the work. He is also indebted to Dr. S. C. Garrison for the privilege of using in the study certain data which the latter had collected. rn . | i he may } f ¥ i , Na Roa \ wont i ea? bie PICT LE e pete as H NAA \ Af a jaa i TAN / , } whi A NMiniet ers. : aR ty ! te re LPN a be Bal ait ; : Ay AVN ae ad at aR linear tM py ae ee tL Y) : ‘ y ; t: Pesta eeine co a ’ at at < PAS aa Gal OLDE fae eh cas . Fer) As AUJAN EA Pa Gus tt aS ay ae <3, Oa EX Re ata) Sait ‘ oF : 4e) : 2 ates ‘ - Pat - y 7 _, ‘= | bane avi Pao PT Ee a aoe 4 MMV GH arent) Paper ar le kl / ie {/ Pes ' ’ * ‘ { att wpiiehnde b ftyis +1 . a AA Ca eS | ' 5 Roe : bet SAS f - ae “4 Aye : ‘ evils ; , Reb .apae > & ; ft \ "aT \ ( verre. cciy mt % eet } barat NiTh one td i) a Ay : "3 i re APR TERE ACTA ty hes * Ou! ma a e 7 ie, y FP), CONTENTS PAGE Be PT LON ce eersgialie ura alt anh delat Maen aimee betes vii TAL Cte POOA TS. See ye at, oo ee ee oe ee 1 110, VOUS G4 DLO Peed Spi pee teen Beppe ees URL her nL Sean el a PSE 11 III. EXPERIMENTAL MATERIAL AND PROCE- UNL AR Ei ge 8 ele eh en ga 12 ote pe U ET CELE sO Sate ius EIR 1 A 12 Pee TSeAN I CL EST) PROCEDURE: 250) '5 oteialgyo le atade als 12 RE ET DCA TAY BPS at A RE OE eg RS A 16 PAmMRELVABILITY OF VBASURES IG onc )i sla hive deata ase 16 eT ICAL UREAT MENT Judd says of these findings: ‘‘ These figures seem to emphasize the fact that good readers are usually not slow and poor readers are usually not fast. . . . For the purpose of this survey the general fact that high rate and good quality are commonly related, and that low rate and poor quality are commonly related is of great importance ” (11, pp. 154, 155). Since a contingency coefficient of only .22 by Yule’s formula RELATION OF RATE OF RESPONSE TO INTELLIGENCE 3 (21, p. 65) is obtained from Dr. Judd’s data, it is not clear just how he arrived at such conclusions. In the same year King (12) published results of 93 university students in reading. Though his method is different from that used in the Cleveland survey, the results are practically identical. He finds a Pearson r of —.07 between rate and comprehension. His method probably placed the slow readers at some advantage over the rapid. He says: “ There are all degrees of difference among individuals in rate of reading. These differences are doubtless due to both innate and acquired factors. Some people react slowly; others move quickly. There is no doubt also but that there are varying degrees of comprehension associated with each differing rate of reaction, 1.e., the more slowly reacting group includes some who are keen to comprehend and some who are quite dull, with all gradations between; and the quick reactors also include various degrees and types, such as the keen, the dull, the thorough, and the superficial ” (12, p. 830). McCall, in 1916,(14) first raised definitely, so far as the writer has been able to find, the question of the probable desirability of “speed ”’ tests and “ power ”’ tests as measures of mental ability. ““ By ‘ power test’ we mean,” he says, “ one that contains units sufficiently difficult to discover the maximal ability of the person or persons being measured.” Binet-Simon scale and Trabue completion test are cited as examples. It should be added that any time limit in a power test should allow opportunity for all to reach the limit of their abilities. Fighty-eight boys and girls in the grammar grades were given a series of speed tests by McCall: Cancellation, handwriting, addition, and looking up and copying addresses from a directory ; and a series of power tests: Thorndike’s Visual Vocabulary, Trabue’s Completion, Thorndike’s Reading Scale Alpha, arith- metic problems, and Thorndike’s Omnibus tests I A and II A. Teachers’ ranks and school marks, composite of all tests, and the average of these were used in turn as criteria of mental ability. As the correlations with each of the three criteria are so nearly the same the average will probably give the best view of the facts. The average correlation of each test with the three criteria 4 J. A. HIGHSMITH follows: Omnibus, .80; completion, .78; teachers’ rank, .75; school marks, .73; reading, .67; arithmetic, .61; visual vocabu- lary, .60; copying addresses, .39; addition, .27; handwriting, 12; Cancelling A’s, S’s, 2’s, and 3’s, —.03, —.06, —.23, and —.23, respectively; and age, —.25. The average correlation of the so-called “ power tests’’ from this group-is .69 and for the ~ “speed tests,’ .03. The average of reliability coefficients for these two groups is .52 for the power tests and .95 for the speed tests, and for school marks, teachers’ ranks, and composite, .88. McCall says that “some of the factors which make for high reliability coefficients are: that the function tested be narrow; that the time spent in testing be long; that the test material and experimental technique for the two tests be identical; and that there be no large variation in the conditions of the subjects.” “To sum up the entire discussion,” he says, “ the power tests give a much higher correlation with mental ability than do the speed tests; and this is true whether average scores or improve-— ment is used as the measure of the speed tests” (14, p. 52). Anderson presents data (1) showing the average rate of mental association for various groups in responding to words presented. Fifteen eight-year-olds averaged 2.6 seconds with quartile devia- tion of .8. Twenty-five ten-year-olds averaged 2.3 + .7, twelve- year-olds, 1.7 + .4; fourteen-year-olds, 1.6 + .6; and adults, 1.5.3. While he thinks the type of answer given has some relation to intelligence, he found no correlation between speed of association and intelligence. , Terman (18) found a correlation of .535 between total words named in three minutes and M. A., and .52 between total words. named in one minute and M. A. The correlatioti of the first and last minute was .58. Terman and his students examined 818 individuals with Stan- ford-Binet and the Army Alpha test and found correlations which averaged .78, ranging from .58 to .87. The groups included were heterogeneous. No attempt, however, was made to eliminate the-ame factor (( 1ainaasz )’ Gates (8) studied the rate of reading in its relation to compre- hension of what was read, and to general intelligence. He used RELATION OF RATE OF RESPONSE TO INTELLIGENCE 5 the composite of several tests in rate, comprehension, and group intelligence measures, and found the following correlations: r SelB, StailOuae mre With ate Mths occ. 6 ¢ «chee le mena tte vo) a ee rol .30 Stantord avi wAacviti, Gomprehension,......ase eee ess one 34 24 SPantard Mi. Me With ITECHONS \...: «2s acsenebancne chase « ae .20 Group iintellizence with Ratel.. 06. 200. Fe Saw ee 64 17 Group Intelligence with Comprehension................ .69 .10 Group Intelligence with Directions...............0ee00: .61 AZ RAL Omni ILI CLIO NS ite iieverccs sis's Jessi, «ise 6'el hv seve ermal eteiataeias .79 14 Rater wits GCOMPrENensiONi shee scids ano s's'hg en etedomiasiaet .84 .08 Ditections with Comprehension). .00 J. eee eee ape .78 .14 In the interpretation of these correlations it is well to keep in mind that the reading comprehension tests were in some cases timed tests. ‘These tests were those of Brown, of Courtis, of Monroe, of Thorndike-McCall, and the Directions test. Of these the Courtis, Monroe, and Directions tests are definitely timed, while the Brown and Thorndike-McCall allow ample time for a child to give evidence of what he has comprehended. We should expect the rate element to enter into the reading comprehension score to a considerable degree. The correlations tend to show this to be the case. Concerning the reading rate and comprehension correlations Gates says: “ The correlations with the composite of group intel- ligence tests is higher than with the Stanford-Binet and those are about as high in the lower as in the higher grades. Both of these facts might be explained by the greater demands of the group tests on reading, which are rather uniformly stable in the various grades, but this explanation is in no way defensible by our data” (8, p. 459). Root (16) has recently given us the results of a large number of correlations between group mental tests and the Stanford-Binet and between various group tests themselves. About 600 children were tested. The median of 67 grade correlations between Stanford-Binet and group tests is .66, P.E. .077. The median _ of 17 grade correlations between group tests is .765, P.E. .05. The correlations obtained when all children are taken together irrespective of grade are .80 and .88, respectively. These last show the influence of age upon the correlations. The National 6 J. A. HIGHSMITH Intelligence test gives a median grade correlation with Binet of .65 with range of .49 to .79. Root points out that one cause of the variations between the Binet and group tests is probably that “the Binet is largely inde- pendent of the element of time; mass tests must of necessity rest on a time basis. We do not know to what extent different sub- jects are benefited in one case and injured in the other, or vice versa”’ (16, p. 292). Gates (9) shows the relation of reading ability to intelligence as determined by three types of intelligence measures: Stanford- Binet mental age, verbal group intelligence score, and non-verbal group intelligence score. The correlation of reading ability with Stanford M. A. was .49, M. D. .16; with verbal group test .71, M. D. .06; and with non-verbal group test .20, M. D. .07. We are by no means justified in saying that the difference between the verbal and non-verbal tests is due to the verbal element. The very fact of a test’s non-verbalness means that a different kind of task is to be performed, which, while differing from the verbal in respect to the verbal element, also differs in respect to the nature of the tasks to be performed. Freeman (6) cites a study in which the correlation between the score on the Burt reasoning test and the time required to — complete the test was zero. Freeman found the correlation between the speed, as determined by the order of finishing, and the quality of performance as measured by two tests on subject matter of class. One hundred fourteen, largely graduate students, were tested with a multiple- answer and a completion test constructed to measure content of mental test course. The correlation between the order of finishing in the two tests was .50+.05. The two tests correlated .55 + .047. Freeman says: ‘ Since the content of the two tests was in part different this indicates again a fair degree of relia- bility.’ The correlations between the rate of work as determined by the order of finishing and the quality were —.13 + .07 and —.12 + .07, respectively, for the two tests. So he concludes that, “There may be real differences in the quality of work RELATION OF RATE OF RESPONSE TO INTELLIGENCE 7 of which one is capable which are independent of speed of performance ”’ (6, p. 88). Gates, 1922,(10) found an average correlation of .55 between Stanford M. A. and the mean of several group tests. The average of the correlation between M. A. and the different group tests was 46, S. D. 15. The National Intelligence tests, forms A and B combined, with Stanford M. A., gave .46, .52, and .58 for grades four, five, and six. Garrison (7) obtained similar results with 158 cases in grades four to eight with the Otis test. The average of the correlations between Stanford Revision and Otis was .48, with a range of 25 tOn. 37% So far as the writer has found, only two studies have been miade directly upon the relation of speed and quality in intelligence tests now in general use. The first of these was in connection with the standardization of the Army Alpha scale (15). Five hundred ten recruits were given the test. When time was called on a test the recruits were asked to draw a line to show how far they had gone. They were then instructed to go on with the same test, but not to go back of the line to correct anything, until time was again called. The time allowed the second part was the same as for the first. Each test was given in the same manner. In this way it was possible to determine what a person’s score was for the two time limits, t.e., single time and double time. The purpose of this comparison was to determine whether there was any marked change in relative position under double time as compared with single time. The correlation obtained between single and double time scores was .965. This means, of course, that there is but little change in relative position due to doubling the time. More instructive, probably, than the coefficient of correlation is the analysis of the data to show the per cent of individuals at each level of the various tests in single time who gained in double time. The conclusions from this analysis are given as follows: ‘‘ We might say, therefore, in the case of these tests, that they are neither principally ‘speed’ tests nor ‘ power’ tests, but tend to show the characteristics of a ‘ power’ test more at the 8 J. A. HIGHSMITH low levels than they do at the high levels. The high frequencies of persons gaining at the upper levels (often 100 per cent) indi- cate that for the people making high scores in single time the ‘speed’ element is predominant. In the middle and lower ranges the ‘power’ element is more important. Many persons do not gain in the additional time. It can hardly be said, however, that at these levels the ‘power’ factor is ever so important as is ‘speed’ ” (15, p. 419). The second study directly concerned with the speed-quality relationship in intelligence tests was recently reported (17) by Ruch and Koerth. From a number of freshmen who had previ- ously been ranked on the basis of Thorndike’s, Morgan’s, and the Iowa Comprehension tests combined, were selected the lowest: and highest deciles, 70 in the low group and 52 in the high. These were given the Army Alpha Form 7 in such a manner as to reveal their scores for single, double, and unlimited time. A correlation of .966 was obtained between single and double time and .945 between single and unlimited. This leads the writers to say that, ‘““The agreements between the correlation of single and double time with that reported in the Army figures is striking.” Apparently these writers have overlooked the fact that correla- tions are much affected by the degree of heterogeneity of the - subjects tested. They have obtained these correlations from highly selected data, i.e., from the extreme deciles of the original distribu- tion. In such a selection they leave out of account those deciles where variation is least or where differences are smallest, 7.e., those nearer the central tendency. A formula for evaluating a coefficient of correlation obtained under such conditions has not come to the notice of the writer. But the fallacy of the correla- tions presented by these writers may be shown to some degree by the simple process of correlating the measures for the two deciles separately. That is, we may take the distribution given by Ruch and Koerth for the low group in single and double time and on the basis of that find the degree of correlation existing when the high group is left out of account. Then we may do the same for the high group with the low group left out. This gives us a correlation of .86 between single and double time for the low RELATION OF RATE. OF RESPONSE TO INTELLIGENCE 9 group and of .71 for the high groups. And instead of the one correlation of .945 between single and unlimited time, according to Ruch and Koerth, we get .76 for the low group and .65 for the high when each group is taken separately. Even the coefficients resulting from correlating the deciles separately are much too high, for we are dealing in such cases with the extremes where differences are greatest. If, for example, we take a probability curve of 3.0 sigma we find that either extreme decile extends over 1.73 sigma on the base line. The decile about the median, on the other hand, is limited to .25 sigma. This means that the extreme deciles have seven times the range of ability that is found in the middle decile. Hence it is much more difficult to discriminate ability in the middle decile than it is in the extremes. Since errors in ranking tend to lower correla- tions, it is evident that the middle decile, with its greatest suscepti- bility to error would give the lowest correlation. That being true, it is evident further that the correlations based upon the extreme deciles taken separately would be considerably reduced if the deciles more susceptible to error were included. Evidence that an obtained correlation may be in excess of the representative correlation is definitely shown in another way. If, for example, we take all of the children of a school system without regard to age or grade placement and measure them with two intelligence tests we get a correlation out of proportion to the relationship which actually exists among children of the same age or grade. Root (16) shows correlations among a large number of tests worked out according to each arrangement. He found the average of the grade correlations between group mental tests and Stanford-Binet to be .66; and between various group mental tests themselves, .76. But when he disregards the grading the correlations are, respectively, .80 and .875. Garrison and Tip- pett (7) made a similar study of 158 children in grades four to eight, inclusive. They found an average grade correlation between the Otis Advanced Examination and Stanford-Binet of .48. When the five grades were thrown together in a single correlation table the result was a correlation of .75. This fact of heterogeneity is in some measure responsible, 10 J. A. HIGHSMITH probably, for the high correlation, .965, found between single and double time in the Army. In 10 of the 13 companies used in the Army study no men had been segregated for the beta exam- ination. The scores for single time ranged from zero to 190. It is of considerable importance, then, to know the nature of the series between which the correlation is obtained. A high correlation in human traits means much or little, depending upon the homogeneity of the group from which the correlation is obtained. Summary of Historical Sketch 1. There is a comparatively low correlation between rate of performing simple tasks and intelligence. The correlations range from .00 (Brown) to .61 (Burt). Most of these fall around .30 to .40 (Wyatt, Brown, McCall). 2. There is a high correlation between rates of performing different simple tasks. Brown gets .80 to .82 and Gates .95. 3. Individuals show considerable differences in rate of response to words presented (Anderson). | 4. Group tests and Binet-Simon tests give correlations averag- ing about .55 to .65 (Root, Gates, Garrison). 5. In all cases linguistic tests give higher correlations with intelligence than do non-linguistic (Gates). 6. The Army experiment and the Ruch and Koerth study do not give sufficient evidence of the closeness of the relation assumed in timed tests between rate of response and intelligence. It. PROBLEM A survey of the literature on mental testing reveals the fact that while the time factor in human reactions has been worked at from many points of view, its relation to general capacity is by no means clear. Are we justified, for example, in assuming that the amount a person can do in a definite length of time is the best measure of his ultimate learning capacity? Are we yet prepared to say that the number of items, such as are found in group tests, correctly responded to in a specified time, is the best measure of a person’s mental capacity? Or, to put it in another way, do the results of studies demonstrate a sufficient uniformity in rate of work among people that its influence upon the measure of their mental power may be neglected? Is there even an approximately perfect correlation between profoundness or sagacity and speed? The purpose of this study is to investigate the relation of rate of response to.general intelligence. We shall attempt to give at least a partial answer to such questions as: To what extent is an individual’s rate of response constant for different kinds of mate- rial responded to? Does rate of response vary with material of different levels of difficulty, and is this difference constant for various individuals? To what extent is intelligence a question of rate of response? Can rate of response in linguistic and non- linguistic materials be weighted with time-limited mental tests so as to improve the correlation with our criterion? This study differs essentially from the other investigations of this problem in that we are employing separate measures for rate of response and for general intelligence, and a third measure combining the factors measured by the other two. Other inves- tigators have used only a single timed test, under different time limits, and, too, without attempting to isolate the rate factor. Ill. EXPERIMENTAL MATERIAL AND PROCEDURE A. Susyects. This study is based upon the responses of 87 boys and girls of the Peabody Demonstration School to the tests described below. Twenty-nine were in the fifth grade, 30 in the sixth, and 28 in the seventh. The median age and 1.Q. were for fifth grade 11.0 and 109, for sixth 12.3 and 109, and for seventh 13.0 and 112, respectively. The complete distributions of chronological and mental ages by grades are given below: DIsTRIBUTION OF AGES IN YEARS BY GRADES Grade 8 85 9 95 10 10.5 11 11.5 12 125 13 13.5 14 14.5 Total Bi Si) meio J avau2o Ro nat Bit ork Siti. geben meas NGAGE HS yee daa Bie Gas Sha), i ae re Tes 8 OS SE OU OE oth he a 1h a Total ui) Sees tiy, Wee (08 MOeds / 0S a8 ig. oa eee en DistriBUTION oF I. Q.’s By GRADES Grade 70 80 90 100 110 120 130 140 150 160 Total Slit des. wtrand tron yg. 12 ddOeedld one eeiieh- ee) Sane aaa Fa geht, caval Le aR 6 iano be Sik Oe 30 TEP SIN YAAV SAN EDR TIT ON See NL) a i 28 Polaris we TL AH LBA SE ates MG a eae Standardized tests are regularly administered in considerable numbers during the year to every grade, so that tests of the kinds used in this investigation are by no means.a novelty to those tested. Their reactions to tests are, therefore, probably more representative of their customary modes of doing tasks. B. Tests AND Test Procepure. The general plan of the inves- tigation was to employ three sets of measures, including one ‘set as nearly free from the time factor as possible, one set including both the time factor and difficulty, and a third set as nearly as possible purely speed tests. It was thought best to restrict the speed tests to such material as is commonly found in intelligence tests, so that there might be good reason for assuming that some RELATION OF RATE OF RESPONSE TO INTELLIGENCE 13 of the same mental functions were employed with the three kinds of data. The tests may be considered under three heads, the criterion, the group mental ability tests, and the rate tests. A. Criterion. The criterion of intelligence used is the average mental age as determined by the Stanford-Binet. For three years, and in some cases four, the pupils used in this study have been tested annually by the Stanford-Binet scale, except for such as have been here a shorter time than that. Fifty-eight of the 87 have had three or more Stanford-Binet tests. Twenty-four have had one test and five have had two. The tests were given under standard conditions, about half of them by Dr. S. C. Garrison and the other half by graduate students in psychology trained in administering the Stanford-Binet scale. It is probably safe to say that the Stanford-Binet test is as nearly independent of the time element as any test to be found. In only 17 of the 90 tests composing this scale are subjects limited in response time; and in practically all of these the time limits provide ample time for those who can do the tests. Only in one test, the word-naming test, is the time factor very similar to that in group tests.’ In none of the tests is there a premium on speed within the limits set. A solution of the code test, for example, in one minute gets no more credit than a solution in five minutes. We may say, therefore, that whatever else this test may measure it does not to any appreciable extent measure speed of response. B. Group Mental Ability Test. From the point of view of the purpose for which this test was to be used there was no basis for _ determining its suitability. Probably any of the better group tests of mental ability would have served. The National Intel- ligence test was, however, especially adapted to the ages we were employing. Form 1, scales A and B, was given and scored according to the standard directions furnished by the authors of the tests. This test was given by a trained examiner under the direction of Dr. Garrison. C. Rate Tests. The aim was to get material similar to that found in group tests and at the same time easy enough to present a minimum of difficulty to the groups of pupils we were using: 14 J. A. HIGHSMITH Further, it was desirable that both linguistic and non-linguistic elements appear in these rate tests. 1. Rate tests with linguistic element: To meet this require- ment, and another to be mentioned later, the Pressey Intermediate Classification test, for grades 3-6, and the Woodworth-Wells Easy Directions tests were selected. The Pressey test consists of a hundred items to be responded to by the multiple choice method. These hundred items were divided into three parts. The division and time limit for work on each were so determined that no individual could finish. Each part had the same time limit, 100 seconds, and this was constant for each grade, as was the method of giving and scoring the tests. The Pressey directions were followed in the main, with such slight modifications as the above divisions made necessary. Each grade was tested separately as a group. The Woodworth-Wells easy directions tests are fully described by the authors (19). The following brief statement summarizes the main points: “ The conditions which it was sought to meet in the test material are: (1) that the motor responses should be very simple and quickly performed; (2) that the instructions should be very simple, but varied; and (3) that the instructions should be as concise as possible, in order that reading time might not be the determining factor.” The two sheets of this test were given at different times. Each sheet had a time limit of 50 seconds for the 20 items. 2. Rate tests of non-linguistic type: The Kingsbury Primary Group Intelligence Scale, Form A, for grades one to four, and the Pressey Primary Classification, Form A, for grades one to two, were selected for this type of rate test. The Kingsbury scale presents various forms to be dealt with. Test 1 was omitted because it was not of the time limit kind. Test 2 is an “‘ opposite’ test; test 3, “completion,” and test 4, ‘form,’ and they contain 14, 12, and 12 items, respectively, or a total of 38 points. iinsic ce samoartcts 12/5) S04.) W238 127.0 6 1| 6.8 113.1 140.5 133.9 |15.7 |49.5 |193.7|109.5 |144.75 133.75. A adh avrg ak tes at La 3,09) 2.24) 2.5 7 re PVT Tae 14) 9 46h S201 (269!) Gc OF a 8.63] 8.80} 7.96) tables which follow. For example, it may be that the low cor- relation between the Stanford-Binet test and the National Intel- ligence test is due to the small P.E. of the intelligence of the group. The Pearson Product-Moment Method was used in all correla- tions. We have assumed linearity in the calculation of all correiations. All arrays have presented, by inspection, a very close approxima- tion to linearity. | RELATION OF RATE OF RESPONSE TO INTELLIGENCE 19 TABLE II j Suowine SixtH Grape Raw Scores 1 ' 1 2 3 4 5 6 % 8 9 10 Intermedi Te: | genie B.-S ntermediate , : o- | sey ings- s=S. 5 ic ahication Directions tal \Priin- aire Total|N.1.T. 7h B.-S.| C.A. ary Cases sf II III | To. | ‘“g” |] dot | To. | 1&2 | To. | To. | 4&5 |A&B} 1.Q. |M.A.| Mo. asi) 13 9 36 6 7 13 49 |43 19 62 235 91 134 |148 Bie 22 18 68 9 18 27 95 |48 18 66 292 |125 177 —-|142 iad. 13 12 39 7 9 16 55 134 9 48 233 97 152 |157 ES 12 13 46 ie 15 32 78 (48 27 75 288 97 142 1/147 D.. 17 10 43 9 11 20 63 146 19 65 269 {116 169 |146 “Uae 13 8 40 8 9 17 57 144 15 59 192 91 124 {136 Bas. 8 6 26 5 7 12 38 130 10 40 191 89 133 |149 Ss. . 26 22 78 12 14 26 104 |53 23 76 336 |1384 173 |129 eo... 22 16 64 14 14 28 92 |46 18 64 287 |136 167 |123 HO. . 14 12 42 8 9 17 59 153 19 72 266," |112 160 |143 a1... 5 5 23 7 9 16 39 139 17 56 215 |104 157 {151 AZ... 16 13 47 9 10 19 66 |47 14 61 314 |135 196 |145 3... 10 11 38 7 8 15 53 |49 25 74 283 {106 160 /151 14... 5 v4 22 6 6 12 34 144 19 63 152 86 136 |155 BS... 16 11 44 8 9 17 61 |41 22 63 298 {108 133 |123 HG... 16 13 44 9 9 18 62 145 15 60 291 NES 1806 {149 es... 9 7 31 7 9 16 47 |36 13 49 173 82 137 |167 8... 16 13 47 8 i 15 62 |45 23 68 200 14112. 156 |139 H9... 12 10 38 8 12 20 58 |41 14 55 242 {113 159 |141 ZO... 8 14 33 6 8 14 47 |46 8 54 192 78 123 |158 Bl. . 15 10 46 9 11 20 66 |43 19 62 276° |107 157 {147 22.. 17 15) 44 8 9 Ay 61 {39 18 Arg 258 {110 171 +|156 A 12 15 44 6 9 15 59 |41 18 59 244 99 145 |146 2 15 14 48 8 9 17 65 |46 16 62 327 11380 183 1141 17 14 50 13 14 27 77 =\44 19 63 255. 1115 155. |135 9 7 30 ri 9 16 46 |46 ‘ie 63 232 81 121 |149 BE sree wg 14 13 45 6 9 15 60 {41 15 56 289 |114 154 1/135 16 16 52 10 14 24 76 149 24 73 289 |121 184 |152 13 9 39 6 8 14 73 \49 15 64 190 98 118 {120 17 15 56 14 9 23 79 = =|57 31 88 278 |110 166 {151 “cot ee 17.5 {14.5 |11.5 |44.4 | 8.43) 9.67]17.33/60.66/45.5 |18.5 |62.9 |264.7/110.0 |157 |145.4 lo _...-| 3.0 | 3.13) 2.48] 1.75) 1.87) 1.87) 3.45)11.1 | 3.52) 3.38] 6.25] 30.5) 9.17) 13.5] 6.85 The absence of correlation between Stanford-Binet and the National Intelligence test in fifth grade is probably due in part, as has been said, to the homogeneity of the group and in part to the presence of certain special reading difficulties in this grade. The writer has worked with this group in an attempt to diagnose cases of reading difficulty. The most pronounced case is that of number 25. This subject was unable to read a second reader acceptably though his I.Q.’s as determined by four yearly Stanford-Binet measures were 114, 113, 117, and 111 for the respective years. Other cases less marked are numbers 22 and 28. It is not surprising that the highest correlations are between tests that involve the language factor. And between tests involv- sing such language factors and those which do not, the coefficients do not equal four times their P.E. 20 J. A. HIGHSMITH St SER TIL alata iat: TABLE III SHOWING SEVENTH GRADE RAw SCORES 1 2 3 4 5 6 vi 8 _ rid pe At B.S Intermediate : c o- | sey ings- a- _: Classification er tal |Prim-| bury Total tional} Aver- MEA ary age Cases I II TIT! | Ton] es) dot) Tos t&2)| “Ta: 4&5 |Total| 1.Q. | Mo. dE eee ea eee IE 16 12 9 37 8 9 17 54 138 25 63 245 7105 158 De ee StS aH 22 17 15 54 12 i 24 785 Mae, 14 51 3385 1140 219 3 24 19 13 56 11 15 26 82 146 25 71 273 |140 217 4 28 22 20 70 13 15 28 9& {45 217 166 316 {116 176 Dad he Oe EUS Tae 16 12 12 40 8 9 17 67 132 20 §2 228 99 150 6 18 15 12 45 8 11 19 64 1/38 iby 53 298 |132 181 Tin eae. Ree ee ee 27 19 18 64 14 nig 25 89 {48 18 66 303 116 175 S.nss. Se des ee 2h 16 17 54 11 14 25 79 46 26 72 309 117 183 O Lae: elt oe ee 16 16 10 42 6 It Lig 59 =|41 22 63 247 |101 167 1O:S3 Pee 16 12 9 37 5 8 13 50 {33 19 52 274 {113 184 11, 4a Sr ae ee 17 9 8 34 7p 9 16 ; 60. 137 aia 48 2240 ails 177 i 12 es hek RO Oa 14 10 8 32 6 7 13 145 |21 12 33 257 $115 178 A 13 19 18 14 51 10 9 i9 70 {41 22 63 273 1104 174 Ww 14 ee ic Beals aloe 13 15 12 40 9 11 20 60 |44 19 63 282 |111- 165 15 16 12 14 42 a 9 20 62 {28 18 46 296 {i160 226 16 SA yes Gee 17 12 12 41 9 ff 16 57 |54 28 82 282 3118 170 17, SAaAKo Seietee eeuae 18 18 16 52 12 11 123 75 \49 23 72 303 {101 138 18 16 12 12 AQ 8 11 19 59. 52 23 75 245 {103 161 19. S:iesas 4 igsavereckexcee he vik 19 17 57 14 12 26 83 |47 21 68 337 = |113 191 20 20 15 14 49 9 11 20 69 |48 28 76 296 |104 175 i DT 5 rane Soka ese 20 14 10 44 5 12 17 61 |49 26 75 255 95 161 PTT a a Sin eh 3 24 21 14 59 13 13 26 85 |42 26 68 290 {121 214 23. Bicashe green aystecal| Lae 20 13 50 10 9 19 69 {39 20 59 287 |102 160 2A ea eee mies tise Wee 28 25 19 72 16 15 31 103 |43 20 63 371 {109 177 25 25 22 16 63 14 18 32 95 |43 27 70 822° {115 171 | 26 i8 15 12 45 8 11 19 64 42 20 62 246 |107 162 | 27 22 18 16 56 12 14 26 82 |50 25 75 288 |107 162 uy 28 25 22 14 61 14 12 26 87 |46 29 75 312 |114 175 a Med aaa iiate a6 sche 19.0 |16.5 [14.0 |50.0 |10.5 |11.6 |20.3 | 68.5]/43.5 |22.0 |65.0 |287.5)113.3 |175.7|154 BiB psiian ue tists oad: 2.82 2.72| 2.24) 7.12] 2.02) 1.72] 3.39) 10.5) 4.84] 3,17 ee 23.4) 9.34) 1 Turning to the correlation table (Table 1V), we find that the only significant coefficients are those between tests involving lan- guage factors. It is evident also from the partial and multiple correlations that the relation between the National Intelligence test and the Linguistic Rate tests is independent of factors meas- ured by the Stanford-Binet tests. 724==.80, and when the Stanford-Binet measures are combined in the best way with the Linguistic Rate tests the coefficient is not materially changed. Tangs ee be 05! The fact that most of the correlation coefficients fail to meet the requirement for significance of four times their P.E. limits us considerably in possible deductions. However, we are still no doubt safe in challenging as a safe measure of intelligence in the lower grades an instrument involving a large linguistic element and which depends upon scores made under time limits. For we 3 q aad utes. = RELATION OF RATE OF RESPONSE TO INTELLIGENCE 21 TABLE IV SHowinc Tora, ParTriaL, AND MULTIPLE CORRELATIONS, AND PROBABLE Errors AMONG Various Factors! For THE FirtTH GRADE 12 08+.12 | 13 324.11 | 14 .18+.12 | 23 at Wits ing Ie t243 0384.13 | 13.2 gla:.17 | 14.2 5206 125). 231 wh Oicte Le 12.4 = Ol seb) 18.4 28 +.12 | 14.3 11.12 | 23.4 =.06--.13 12.34 .08+.13 | 13.24 28 t.12 | 14.23 L512), 23.14 an Ait sh 24 80+.04 | 34 262.11) 1.23 .82+.11 | 2.13 pLSiets ba 24.1 80+.05 | 34.1 2b aby LQ y He Ie 214.12 | 2.14 81.04 24.3 80+.05 ! 34.2 204.12 | 1.34 .383+.10 | 2.34 .81+.04 24.13 80+.05 | 34.12 154.12 | 1.234 8524.16 | 2.134 .81+.04 1 The factors correlated are designated by numbers as follows: 1. Criterion. Stanford-Binet tests. 2. National Intelligence test. 8. Non-linguistic rate tests. Pressey Primary and Kingsbury combined. 4, Linguistic rate tests. Intermediate Classification and Directions combined. cannot overlook the fact that the non-time-limit Stanford-Binet does not correlate significantly with any timed test, while there is a high correlation between timed tests with the linguistic element. It is impossible to say to what extent the low. correlations in this grade are due to the narrow range of I[.Q.’s. Below are given the probable errors of the distribution of the National raw scores, of the average I.Q.’s, and of the Mental Ages for the three grades. PROBABLE Error oF DISTRIBUTION Grade National Aye tsQ): M.A. RC Sete hte frets rey hee diate tee 24.75 8.80 8.80 (Oe ois) ag 1d cece iol GRE BOLO EEE Sc Totes Aer sia Tce 30.50 9.17 13.50 RT Rae oe RSE eye rrihee MU] Laivhay'e Colao oc « 23.40 9.34 13.80 Especially in mental age distribution the fifth grade is much narrower than the others. The lower P.E. for the National in the seventh grade is no doubt due in part to the restriction in amount of scores imposed by the limits of the tests. It is a fact, further, that the non-linguistic rate test does not contribute to the correlation of the National with the Stanford, since v2.14 == .81 and 72.134 = .81. Table V shows the correlations from the sixth grade data. All coefficients of the zero order except one (713) meet the require- ments for significance of four times the P.E. The influence of the rate factor is shown clearly in the multiple correlations. The correlations of all the factors measured by the Stanford-Binet with all the factors measured by the National 22 J. A, HIGHSMITH Intelligence test is .76. The same coefficient is obtained for the National Intelligence test with the Linguistic Rate tests. The partial correlations, however, show that each of these three tests possesses different factors or the same factors in different degrees. Since 7ve14==.87 and 72.134 == .88, it is evident that the non-: linguistic rate test does not influence materially the correlation TABLE V SHowine Torar,. PARTIAL, AND MULTIPLE CORRELATIONS AND PROBABLE ERRORS AMONG Various Factors! For THE SIXTH GRADE 12.3 73+.06 | 13.2 15+.12 | 14.2 12+.12 | 23.1 47 +10 12.4 65+.07 | 13.4 05+.13 |} 14.3 44+.10 | 23.4 21+.12 12.34 -65+.07 | 13.24 —.12+4.12 | 14.23 —.06+.13 | 24.14 23 +.12 24 76+.05 | 34 56+.09 | 1.23 CEES Ads: 82+.04 24.1 65+.07 | 34.1 48+.09 | 1.24 76.05 | 2.14 87 +.03 24.3 65+.07 | 34.2 274.12 | 1.34 53+.09 | 2.34 77+.05 24.13 534.09 | 34.12 13+.12 | 1.234 77+.05 | 2.134 88 +.03 1 See note under Table IV. between the Stanford and the National. The significant factors. in producing the correlations shown are contained in the criterion, the National, and the linguistic rate tests. The correlations 71.24 == .76 and riz==.76 show that the lin- guistic rate tests do not contribute to the National any factor not already contained in the latter so far as its correlation with the ~ criterion is concerned. The correlation r2s shows further that the National and the linguistic rate tests have much in common. The factors, then, that are measured by the linguistic rate tests are also measured fully by the National. The question as to whether the National and the linguistic rate tests emphasize factors not contained in the criterion leads us to consider the effect of combining the rate tests with the Stanford for the highest correlation with the National. riz.4 == .65 + .07 is an indication of the closeness of the relation between the Stanford-Binet and the National when factors measured by the linguistic rate test are held constant. Now 7r2.11==.87, while ri2==.76. That is, when we correlate the National with the best weighting of the Stanford-Binet and the linguistic rate tests combined we get a coefficient which shows the degree to which the National contains the rate factor independently of any signifi- RELATION OF RATE OF RESPONSE TO INTELLIGENCE 23 cance for intelligence. The lowering of the coefficient by eliminating rate, ri2.4==.65, indicates that rate is a factor in intelligence as contained in our criterion; while 72.14 == .87 shows that we must add rate elements to the Stanford-Binet to offset the excess in the National. The seventh grade (Table VI) presents a somewhat different situation from either of the other two with reference to our TABLE VI SuHow1nG ToTat, ParTIAL, AND MULTIPLE CORRELATIONS AND PROBABLE ERRORS AMONG Various Factors? FoR THE SEVENTH GRADE | 12 34.11 | 13 —,2on. 12 | 14 .22+.12 | 23 18.12 12.3 40+.10 | 13.2 en line Dare ES —,09+.13 | 23.1 .26+.12 12.4 28+.12 | 13.4 —.00 1k | 1413 ,of sb) BD 2334 =, oa) 12,34 18+.12 | 13.24 —.31+.11 | 14.23 .0O84-.13 | 23.14 24.12 24 .80+.04 | 34 .44+.10 | 1.23 .45+.10 | 2.13 -41+.10 24.1 .79+.05 | 34.1 .62+.09 | 1.24 344.11 | 2.14 824.04 24.4% .82+.04 | 34.2 .50+.09 | 1.34 .43+.10 | 2.34 824.04 24.13 79+ .05 | 34.12 .52+.09 | 1.234 .46+.10 | 2,134 5324.04 1 See note under Table IV. problem. The extent to which differences in reading rate and practice in taking tests have entered into these group differences and also into the differences between results from these groups and the results from groups measured by other investigators, is beyond the scope of this study. The results here agree with those of the other two groups in showing that the non-linguistic rate factor plays but little part in the correlations we are considering. They also agree with the results in the other groups in showing a marked correlation between the linguistic rate tests and the National. There is a striking absence of significant correlations between the Stanford-Binet and any of the other measures. The correla- tion of .80 between the National and the linguistic rate is raised to only .83 when the best weighting of the two rate factors is combined with the Stanford-Binet. This fact indicates that the National Intelligence test is a much better measure of such rate factors as are measured by our rate tests than it is of the factors measured by the Stanford-Binet. So far we have observed the influence of the rate factor in the grades separately. We have, also, observed that the different 24 } J. A. HIGHSMITH grades show different degrees of relationship when the same pair of traits is considered. In other words, the results from the three grades are not in absolute agreement in many cases as to the TABLE VII SHow1InGc AVERAGE GRADE CORRELATIONS, PARTIAL AND MULTIPLE CORRELATIONS, AND AverAGE DEVIATIONS AMONG THE Various Factors! For THE THREE GROUPS 12 39.24 | 13 14 .25 | 14 31 12 | 23 .30 16 12.3 39.24 | 13.2 05 .24 | 14.2 14 1) } 23.1 .29 til 12.4 31 .23 | 18.4 —.03 .24 | 14.3 SLES LO i 2a04 —.06 18 12.34 29 1.24 | 13.24 —.05 .28 | 14.23 06 .08 | 23.14 12 15 24 79. .01 } 34 42 .10 | 1.23 51 18 | 2.13 47 .20 24.1 75 =.06 | 34.1 -49 13 | 1.24 .43 21! 2.14 83.02 24.3 76 .07 | 34.2 32, htl2 jel ot .43 .07 | 2.34 80 .02 24.13 71 = .12 | 34.12 ey ieee Wee eae .53 16 | 2.134 84 .03 1 See note under Table IV. degree of relationship existing among the various measures. It is necessary, therefore, that we combine the results from these grades if we are to know what general tendency is representd in our results. Two methods have been employed in bringing together the results from the three tables of correlations already discussed. Table VII shows the average of the grade correlations shown in Tables IV, V, and VI. This table shows the mean tendency in the relationships of the various factors. correlations obtained when we throw together the scores of the 87 TABLE VIII Suowine Torar, PartiaL, AND MuttTipte CoRRELATIONS AMONG Various Factors! FoR ALL GRADES COMBINED 12 . 64 13 .33 14 56 23 .57 12.3 .58 13.2 .05 14.2 .02 23.1 .49 12.4 .35 13.4 .02 14.3 47 23.4 .07 12.34 .35 13.24 .00 14.23 .00 -* 23.14 -O1 24 .88 34 . 62 1.23 . 64 2.13 74 24.1 . 82 34.1 .55 1.24 . 64 2.14 .90 24.3 83 34,2 129 1.34 .56 2.34 .89 24.13 nae 34.12 .30 1.234 64 2.134 .90 1 See note under Table IV. subjects for each test and correlate them as a whole. There are certain advantages peculiar to each method which will be discussed in connection with the two tables. Before entering into a discussion of the meaning of the correla- tions in these two tables, however, some justification should be Table VIII presents the - RELATION OF RATE OF RESPONSE TO INTELLIGENCE 2 given for omitting the age factor from the partial and multiple correlations. There are two reasons for the omission. In the first place, the three groups dealt with here represent highly selected cases in which age correlates from very low to negatively with the other factors. There is a tendency for the younger children in each grade to have the higher mental ages, whereas in unselected cases there would be a substantial positive correlation between chronological age and mental age. If, then, we include the age factor, influenced as it is by selection, in our partial and multiple correlations, we are including a factor whose obtained correlation with the other factors cannot be the true correlation. In the case of the National Intelligence test and chronological age, for example, we find grade correlations of —.23, —.33, and —.06 for the fifth, sixth, and seventh grades, respectively. When, however, the three grades are thrown into a single table the correlation is .28. In the second place, we are concerned here chiefly with the question of the relative significance of speed of work in the measurement of intelligence by means of timed and non-timed tests. If, then, the omission of the age factor should introduce a constant error its influence would be chiefly to change the absolute amounts of the correlations rather than the relative amounts. The correlation between the coefficients in Tables VII and VIII is .95. Our next problem is to study the implications of the combined correlations in order to see what they may contribute to the solution of our problem. We shall confine the discussion chiefly to the implications of the multiple correlations resulting (1) when the criterion is correlated with the best combination of the other factors, and (2) when the National is correlated with the best combination of the other factors; that is, v1.23--n and f2.13--n. ‘We shall present also the correlations from both Tables VII and VIII, so that in connection with the discussion of any point the relative changes in the correlation resulting from the inclusion or exclusion of the various factors may be observed in both tables. It should be kept in mind that Table VII represents the averages 26 J. A. HIGHSMITH of the grade correlations, while Table VIII gives the correlations when the 87 cases are thrown together in the several tests, but without any attempt to control the age factor. Whenever there is a discrepancy between the correlations in the two tables we shall assume that the difference in either case is unreliable. We have also determined the limits within which differences do not equal or exceed four times their probable error. By obtaining the probable error of the difference between coeffi- cients, we find that with a correlation of .30 a difference of .04 is significant and that with a correlation of .80 a difference of .01 is significant. It is also true that correlations of .30 and above are four times their probable error in these tables. It is evident from this table that the linguistic rate test does not measure any factor involved in the criterion which is not measured by the National Intelligence test or the non-linguistic rate tests. This is shown when we combine the different factors by multiple correlation so as to get the highest correlation with the criterion. Table VII Table VIII 1.23 = cot 64 11 .234= pos .64 From this it is seen that practically nothing is contributed by the linguistic rate factor. A correlation of .51 is raised to only .53 by including the linguistic rate factor. That the rate factors, especially the non-linguistic, contribute something to the National as a measure of intelligence is shown by the following correlations: Table VII Table VIII 12 = 39 64 1.23 = a 64 1.34 = 43 .56 11.234 = 93 64 Observe first that a correlation of .39 between the criterion and National is raised to .53 when the two rate factors are combined with the latter. Again, it is notable, in this connection, that the two rate factors when combined correlate with the criterion slightly in excess of the coefficient obtained between the National and the criterion. Next, let us analyze the results when the criterion is combined RELATION OF RATE OF RESPONSE TO INTELLIGENCE 27 with the two rate factors to give the highest correlation with the National Intelligence test. If we examine the correlations below we find that the rate tests and the National Intelligence test measure factors largely independent of those measured by our criterion. Table VII Table VIII 24 = 79 .88 124.1= Vk, .82 When we keep constant the factors measured by the criterion which may be present also in both the National and the rate tests, we find that the correlation resulting is reduced but slightly. And if we combine the criterion with the rate tests for the best correlation with the National the coefficient is but slightly higher than the total correlation between the National and the rate tests. Table VII Table VIII 24 = 79 .88 2.14= .83 -90 These facts seem to justify the conclusion that the factors measured by the rate tests and by the National Intelligence test are to a very large extent identical. To what extent are these factors also identical with factors in our criterion? The following correlations show the relation between the criterion and the National and between the criterion and the rate tests, respectively : Table VII Table VIII 12 = .39 64 11.34 43 .56 These probably mean that there is not a reliable difference between the two correlations. This is but another way of saying that the rate tests and the National maintain about the same degree of relation with the criterion. If we take the two correlations, Table VII Table VIII 12.34 = .80 .89 72.134 84 90 it is evident that the criterion may be left out with but little loss in the degree of relationship. But if we compare the following correlations, 28 J. A. HIGHSMITH Table VII Table VIII 1.34 = 43 .56 11 .234—= Jo 64 we find that the National does contribute slightly factors which are not included in the rate tests. The correlations, Table VII Table VIII 11.34 43 .56 12.34 .80 .89 show the degree of correlation of the criterion and of the National Intelligence test, respectively, with the rate tests combined. The implications of these correlations are, first, that the cri- terion is a relatively poor measure of rate of response ; second, that the National Intelligence test is a relatively good measure of rate of response; and third, that the National is a better measure of what is measured by the rate tests than it is of what is measured by the criterion. It was hoped that by using as rate measures tests of slightly different difficulty we might get data for studying the relation TABLE IX SHowinc THE Meprans, DIFFERENCES OF MEDIANS, AND P.E. or DIFFERENCES FOR THE Various Grapes AND Dirricuttiges. I, II, anp III stanp For THE THREE DEGREES OF DIFFICULTY FROM EASY TO MOST DIFFICULT _ Medians I and II II and Ill Grades ——_ | | I II III | Difference |P.E. of Dif.| Dif. |P.E. of Dif. Bass Heys a Soo ena 12.5 &.4 eee 4.1 744 heh 653 eh, Vis Aedes ent ae e geeee eae b Wy fees oe WSs 9 i Ne 3.0 815 3.0 . 749 AR er LaAE Ae Orch oR oder oels 19.0 } 16.5 | 14.0 2.5 776 2.5 . 698 Médinnt 7205) a etean 16.0 | 13.4 | 13.2 2.6 617 2.2 539 _. ) The medians, differences, and P.E. of differences in the last row are determined from distributions of all the scores in each of the three degrees of difficulty. of rates at different degrees of difficulty to intelligence, and at the same time not seriously interfere with the original require- ment for a rate test. It seems, however, that the slight differ- ences in degrees of difficulty and the very small amount of time given to each different degree have operated to make the correla- tions somewhat inconsistent in some places. We are, neverthe- less, presenting the results and indicating some of the implications which seem to be most justifiable. Table IX is presented especially to show the amount and significance of the differences between the different degrees of RELATION OF RATE OF RESPONSE TO INTELLIGENCE 29 difficulty of the rate tests. Greater degrees of difference in difficulty between these tests would have brought out better the influence of the difficulty factor. Two questions will be considered in connection with these measures of rate at different degrees of difficulty. We wish to TABLE X SHOWING THE AVERAGE OF THE TOTAL, PARTIAL, AND MULTIPLE CORRELATIONS AND AVERAGE DEVIATIONS AMONG Various Factors? 12 ==, 59) (24 12 == )j100 » get 12 aes OOM Vee 13 == ole 40 14 Se Oe Le 15 Ey ane oes 23 === 7 ay OD 24 an OL) LD 25 aa 19) OF 12.3" °==(267) 925 laski Hs) lose et 12.57 == 01 | 740 13.2 e=—.06 an uLO 14,2 =—.07 .12 NGO tee 0 NE be 133420 =.17'')/09 14534 == 0T 19 163) =e LOY Sy, 13.5) \===06., 3138 to. == 05, ..08 LB 4) = 24 oy toe One Oe SOS Me a OOS AS, Li ==.80)) 207 34.5 ==.36 .26 35.4 == .44 .09 AG Shee ole eek 2071 pe Ot ehO ee A see Ch OO Sha) ==) 7450: 20 23.4745 isl Me24 f= A220 TL. 2b) == 60" 214 oak =au79 03 2014 Su 817, 08 2, £6 )=783'97. 01 Dele Gish. se A Aa 6003 §.12° =="80, .03 1.345 —.44 11 3.145 == } 86). 701 4.1385 =.87 .02 ; = .87 .06 1 The factors correlated are designated by numbers as follows: 1. Criterion. Stanford-Binet Tests. 2. National Intelligence Test. 3. Linguistic rate test of least degree of difficulty, I. 4. Linguistic rate test of slightly greater degree of difficulty, IT. 5. Linguistic rate test of greatest degree of difficulty, ITI. consider the influence of the different degrees of difficulty upon the correlations of rate, (1) with the National, and (2) with the criterion. It should first be pointed out, however, that the corre- lation of the rate tests and the National with each other give much more consistent correlations for the three grades than does either with the criterion. This is shown clearly by the average deviations in Table X. The following average coefficients from Table X show that the slight difficulty differences in the rate tests are concerned in the correlation: 123 = 72 124 = .76 125 = .79 This comparison is offered with the reservation that the pos- sible effect of practice upon the correlations in the cases of the second and third degrees of difficulty has not been determined. These correlations were obtained from the following total correlations : 30 J. A. HIGHSMITH Grade 123 124 125 5 793 be) 82 6 654 ere 76 7 723 778 786 In only one case, 724 == .77 in the fifth grade, is there a failure of the correlation between the National and the rate tests to increase with the increase difficulty of the rate test. This excep- tion is no doubt due in part to the narrower range of ability in the fifth grade already pointed out. The fact that this relationship is independent of those factors in the National Intelligence test and in the rate tests which are common to the criterion is shown by the partial correlations, 123.1== .67 r24.1=.71 125.1== .74 This indicates again that the National is primarily a rate test of the linguistic type. It also indicates that a single short rate test gives a very good measure of this rate factor. What is the relation of the rate factor to the criterion? In general, the answer is that the relation is small. The total cor- relations between the criterion and the rate tests of different degrees of difficulty show a low coefficient, .30, A. D. .028. The correlation of the National with the criterion is only slightly improved by combining in the best weighting the National with the rate tests of different degrees of difficulty. In fact, this secures coefficients only slightly higher than are obtained between the criterion and the rate tests of the three degrees of difficulty, leaving the National out. The correlations for these are: 11.23 == .45 r1.24== .42 r1.25 = .50 r1.345—= .44 It should be kept in mind, too, that factors 3, 4, and 5 were measured in 100 seconds each, or a total of five minutes working time. This correlation between rates and criterion is somewhat better than is obtained by the National and the criterion, 7.e., ee oo. V. SUMMARY AND CONCLUSIONS A. Procepure. In this study of the relation of rate of response to intelligence two essential features should be noted: 1. The Tests. Three sets of tests were employed: (a) The criterion is the average of I.Q.’s obtained from yearly tests with Stanford-Binet for periods of from one to four years. (b) The scores from National Intelligence tests A and B are taken as a measure of factors present in the criterion as well as of possible factors which might be independent of the criterion. (c) Lin- guistic and non-linguistic rate tests were employed to measure the possible factors in the National Intelligence test which might be independent of the criterion. These tests provided the basis for studying the rate factor with and without the linguistic ele- ment. One of two linguistic rate tests was composed of three parts differing slightly in difficulty, so as to provide the basis for studying influence of rate at different levels of difficulty upon intelligence measurements. The reliability of the tests was determined in each instance. A fairly high degree of reliability (about .80) was found where the self-correlation method was applicable. The only tests for which a high degree of reliability could not be shown were the non-linguistic rate tests. These gave a correlation of .67 between tests of somewhat different material. 2. Method of Treatment. The correlation method was em- ployed throughout. Pearson’s product-moments method was employed in the zero-order correlations. The formulae as given by Yule were used in the partial and multiple correlations. The reliability of individual coefficients is given, as is also the reliability of the difference between coefficients where differences are important. The data from the three groups of subjects were treated first without disturbing these groupings. The averages of the various correlations, as well as the original coefficients, were tabulated. 32 J. A. HIGHSMITH The data for the three groups were then thrown together and treated as a single array. The results are analyzed for two purposes. The data were studied, first, for the purpose of finding out how important the rate of response factor was in the measurement of intelligence; and second, for the purpose of finding what relation existed between rates based on material of different degrees of difficulty and intelligence. B. CONCLUSIONS: 1. The results of this investigation indicate decidedly that the rate of response to test material is by no means a safe measure of intelligence. 2. They indicate, also, that the National Intelligence test is a much better measure of rate of response than it is of intelligence. 3. The simple linguistic rate tests used in this study are about as good a measure of intelligence as is the National Intelligence test. 4. Rate in linguistic material can be measured much more con- sistently by a short test than rate in non-linguistic material. 5. The non-linguistic rate tests, when added to the National, contribute slightly more to the correlations with the criterion than does the linguistic rate. , 6. The high correlation between rate tests and the National Intelligence test points to a danger in employing the composite of group tests as a criterion by which the validity of a new group test is tested. It may be that this process is increasing the importance of the rate element in group tests at the expense of factors more significant of general intelligence. 10. 11. RELATION OF RATE OF RESPONSE TO INTELLIGENCE 33 BIBLIOGRAPHY . AnpEersoN, M. An Investigation Into the Rate of Mental Association. Jour. Educ. Psychol., 1917, 8, 97 f€. . Binet, A., & Simon, TH. Methods nouvelles pour le diagnostic du niveau intellectuel des anormaux. L’Annee Psychologique, 1905, 11, 191 ff. . Brown, Wm. Some Experimental Results in the Correlation of Mental Activities. Brit. Jour. Psychol., 1910, 3, 296 ff. . Brown, Won., & THompson, Goprrey H. Essentials of Mental Measure- ment, 1920. . Burt, C. Experimental Tests of General Intelligence. Brit. Jour. Psychol., 1909, 3, 94 ff. . FREEMAN, F. N. Notes on the Relation Between Speed and Accuracy or Quality of Work. Jour. Educ. Research, 1923, 7, 87 ff. . Garrison, S. C., & Tippett, J. S. Comparison of the Binet-Simon and Otis Tests. Jour. Educ. Research, 1922, 6, 42 ff. . Gates, A. I. An Experimental and Statistical Study of Reading and Reading Tests. Jour. Educ. Psychol., 1921, 12, 303 ff., 378 ff., 445 ff. . Gates, A. I. A Study of Reading and Spelling with Special Reference to Disability. Jour. Educ. Research, 1922, 6, 12 ff. Gates, A. I. The Correlations of Achievement in School Subjects with Intelligence Tests and Other Variables. Jour. Educ. Psychol., 1922, 139129 ff: Jupp, C. H. Measuring the Work of the Public Schools, 1916. lla. Ketiey, T. L. Statistical Method, 1923. in 13, 14. a3: 16. 17. 18. 19. 20. 21. Kino, I. A Comparison of Slow and Rapid Readers. School and Society, 1916, 4, 830 ff. KUHLMANN, F. A Handbook of Mental Tests, 1922. McCatt, W. A. Correlation of Some Psychological and Educational Measurements. Teachers College Contribution to Education, 1916, No. 79. Psychological Examination in the United States Army. National Academy of Science, Memoirs, 1921, 15. Root, W. T. Correlations Between Binet Tests and Group Tests. Jour. Educ. Psychol., 1922, 13, 286 ff. Rucu, G. M., & Koertno, W. “Power” vs. “Speed” in Army Alpha. Jour. Educ. Psychol., 1923, 14, 193 ff. TERMAN, Lewis M. Some Data on the Binet Test of Naming Words. Jour. Educ. Psychol., 1919, 10, 29 ff. WoopwortH, R. S., & We tts, F. L. Association Tests. Psychol. Monog., 1911, No. 57. Wyatt, STanteEy. The Quantitative Investigation of Higher Mental Processes. Brit. Jour. Psychol., 1913, 6, 109 f€. Yute, G. U. An Introduction to the Theory of Statistics, 1922. 4 i ‘ y Senet: ie ea ibs ) z Ore iT sahictl ey " Tet, ri sb hindi DD he he , m " ena yy AL Iw Mei Hi} eg REAR Chat ik Yi hep un BAN natant alge | Ais ss Radic AE Sabai at? ‘ak ETRE any ot fi we Pe Sa ‘ ¥ ‘ ua he 2 ae a5 nese , kana ahs i ph a } ae st wey a anita Nae fe . ¥ pore hat r oa) & ho, aan ei “tr ied Chg APS 4 Per iyi! 444) 4 tay Si t yo Ieee + 1a, me . ae Madd my ce A eet Ti ae Ay an wAawinsss ihe Chashinse Ne ewe “ ; uf Ra NGS oth AAS Pak (see bx i? ee dea us, ‘ ie Ti a] hms , a iy ait Bd SRE ae mcd } R + P Py q ' M ~ ary * ee vi , te By : 7 i , a v ed : 4A : ‘ ae hee ays | Pratt . Ta Ay eta Bue t: ee) Ac Pye t > ia) TY ‘set | 4 ry) POOR Se: Batt) AD sh) we Mer ome f ; hae WP A Joan ae ; ; . re sire paev ie, SOT) TN f 1a ay ehh: Oe: ane * qT? : \ ty OR Ee mAY qicfet® Low deleliy' i yA Ka¥%s / aye ; at A pene Rak cn A itl ‘ ‘ae ’ fee aes pat by . We Pi aie A : a> in Aye fi 4, ek ; Ag oy ‘. fre d ry iy te ola Ne ig Bia i f ' i : : sn 7 7 peat di ' 4 ae | avi || seuiy ee. 9 tle Le Vib ore ot basa Ate bd et as hs Ae Gy, ts LY | ‘ } - } f ’ ‘ ; ‘new RAT eg al bet re (Hs 7 1 wt pn A sites oe a Phi hs it We bel ck : a co eae t ay Pa ta : % ': “ rr Poet ror i i i i , A ri a 2” tee OF, ya Li co ms & ; io ; ‘on he 4 Lecnd are Y rv iit ‘ ie d ‘ Ls i ven dite 4 4 7 a4 i ; , . 4 ai ‘ = ‘ } are Md q 7 ¥ ] % . ‘ ae yw 2) rie Tor ye uit we bag at \ a a rei) me Os hw i l= | > ne } r - i \ aaa uae to ‘ ha P F chs 4 : a » ns, *huadt ti pts) Van te ipo ty Yr rane i it ay, futte Rh aes Fy f Y J ‘ - ‘ ; HEAT Pe, 4 bMS pe EAM Maines i veel vi i i arty ; a) m4 Whi As a safe Sana ie ¥ iq , ~ fn ce ae he > ‘ ' ‘ a \ F ” ; , ea ' . > 2 ri oa a i i ¥ f ri ) - A; ae 41 : i ‘ene P * J of a a. a ' ’ f ' f y } . 4 : é ays t rary f @ d ; j Aa } i . °) id y i a ig “a . f ae tix Sy ‘4 i ya a Prive) Ne Lee pene) iyi Ses Tare Mal ce ae oe it) en j My Tae Ne ge BF21 .P96 v.34 The influence of tuition in the Princeton Theological Seminary—Speer Library MAM 1 1012 00008 5474