FAMILY ENVIRONMENT AND - * By JOHN NISBET, M.A., B.Ed., Ph.D. Lecturer in Education, University of Aberdeen ABERDEEN, as research workers in not unaware of the fact that environmental other fields have discovered, is a good influences enter into intelligence test scores district for population studies, for its and affect the size of the negative correlation position as the centre of a comparatively between intelligence test score and family isolated region and its mixture of farming, size. T-hey have attempted to allow for the fishing, industry, commerce and administra- influence of environment in two ways, both tion make it in many respects a representa- of which seem to have serious weaknesses. tive urban population. Almost all the Some have tried to measure relevant environ-- children under twelve in the city, in the mental factors, such as parental occupation Scottish tradition, attend local schools, and or overcrowding in the home, and then have the Education Authority is progressive in observed the effect on the negative correla- authorizing and assisting research projects. tion when these environmental factors are The investigations described in this paper held constant. Such a method, however, could not have been carried out without the does not give accurate results if (as is sug- generous facilities afforded by the Education gested in this paper) one of the environ- Authority,t for it was possible in the mental factors is the size of family itself, the administration of tests to achieve an almost amount of contact between adult and child' complete coverage of age-groups of children, and the consequent stimulation of the child's only two small primary departments of verbal development. Again, other investiga- private schools being omitted. tors have made allowance for environment by In this paper a hypothesis is advanced first estimating the future trend of intelli- which to some extent reconciles certain of gence as if intelligence test score were deter- the contradictory results which have emerged mined wholly by heredity, and then halving from recent studies of intelligence and family that estimate, since at least half the variance size. Investigations over the last thirty of intelligence is attributable to heredity. years-there have been at least thirty-three This, however, is not altogether a valid pro- on this problem, among them particularly cedure: for it does not follow that if half the the Scottish Mental Survey with its 70,000 scatter of intelligence is due to heredity, then cases-have established that in any repre- the negative correlation of family size and sentative sample of the population there intelligence is half attributable to hereditary exists a negative correlation between the factors. If a variable X (e.g. intelligence test intelligence test scores of children and the score) is determined by two factors A and B size of the family to which they belong. In so (e.g. hereditary and environmental factors), far as intelligence is inherited this seems to then the correlation of X with any other imply a decline in the level of national variable Y (e.g. family size) may be due to an intelligence; yet two recent studies+ have association of Y with A or with B or with not found any such decline. Those who have both A and B together. attempted to predict the amount of the possible decline in national intelligence are The Hypothesis The hypothesis which is advanced in this *A paper read at a Members' Meeting of the Eugenics paper is that family size has a direct effect Society on January 2ISt, I953. of t The Education Authority is not in any way on the environmental aspect mental responsible for this investigation or its conclusions. development. This hypothesis derives from + Scottish Mental Survey, 1949; Cattell, I950. the view that and words afford a 3' 32TH-- -; THE EUGENICS REVIEW syste-m of symbols which greatly increase the guage development is affected by age of efficiency of abstract . Limitation of associates, children-who associate most with opportunities for verbal development is- adults being' superior. to others: thus only therefore likely to exercise a depressive children show- a striking superiority in influence on ability to score in a test even of , and children from general mental ability. It has been estab- institutions a marked retardation. This lished that the only child enjoys a much finding was based on her own work (1930) on greater verbal development than the child the subject, and was supported by a number from an orphanage, because of greater oppor- of studies reported from Iowa University. A tunities of contact with adults and of conclusion in accordance with these was that acquiring adult . The large family of Smith (I933), that pre-school children use is considered here as an environment midway longer sentences and ask more questions between that of the only child and the when alone with an adult than when in a orphanage. The mere fact of belonging to a group of children. Day (4932) and Davis large family implies restricted contact with (i937) found a retardation in verbal develop- adults and fewer opportunities of acquiring ment among twins; since this retardation adult habits of and thought, a dis- was less marked after school attendance advantage which enters into the intelligence began it was suggested that the inferiority of test performance -,of children from large twins to non-twins was due not to any pre- families. If this hypothesis is correct, if this natal handicap but to their family environ- is a factor in the negative correlation of test ment, in which twins are at first inclined to score and family size-and it is not suggested communicate non-verbally with each other that it is the only factor-then a reduction in and to have less verbal association with the average size of families will mask any adults. In the Scottish Mental Survey decline in national intelligence which may be (Mehrotra and Maxwell, I949), the average occurring. From the point of view of future test score of twins was slightly below that of research one important conclusion would non-twins, a phenomenon which may have result if the present hypothesis were con- been associated with a verbal handicap. /In firmed: in any comparison of the mean view of all these findings, it seemed probable intelligence scores of two generations, such that children from small families would have as is strongly urged in the Report of the Royal better opportunities for language develop- Commission on Population (1949), the advan- ment than children from large families, who tage would tend to be with the generation have relatively less contact with adults and with the smaller family units, and some would thus suffer a slight environmental allowance for this would have to be made. handicap. / "This is the sense in which the phrase Such an influence of family environment on " family environment " is used in the title verbal development could reasonably be of this paper; it is not concerned with expected to affect scores in a verbal test of income or nutrition or the number of rooms intelligence, merely through the words used in the house-though of course these may be in the test.. However, the effect may be more important factors-but only with the con- direct than this: ability to manipulate tact between child and adult, which is on verbal symbols seems to play an important average proportionate to the size of the part in the process of thinking, and particu- family, except that in a large family the first larly in problem-solving. This is the point and the last perhaps have some advantage of view of Terman (1937), who wrote: over the others. (And, in fact, a tendency Language, essentially, is the shorthand of the towards slightly higher test scores among higher thought processes, and the level at which first- and last-born has been observed.) this shorthand functions is one of the most A summary of investigations of the lan- important determinants of the level of the guage development of children is given by processes themselves. (p. 5.) McCarthy (I946), who concluded that lan- This is not by any means a new idea. It FAMILY-ENVIRONMENT 1- 33 appears in the of Dewey (I909), the Gaelic-speaking population of Lewis., Thomson (I924), Ballard (I934), Watts found the average performance of the bilin- ('944) and others, even of widely different guals to be below normal. Christophersen background: for example, Head (I926), (I949) suggested that in the case of mono- - on , noted that speech dis- glots, " after early youth ... a linguistic turbances sometimes impaired ability to perform mental operations which requifed 'crystallization' sets in: their speech the use of symbols. The problem of the rela- organs become, fixed in certain grooves, and tion of language and intelligence is not an their become inseparably linked easy one to put to the test of experiment, with the words of their mother tongue"; since in most cases verbal ability and but in the case of a bilingual person his general mental ability develop normally side " mental processes are not immediately- by side. However, in two situations, those linked with either of his ," a theory' of deaf children and bilinguals, an experi- which " will account for that hesitancy of mental design has been set up by nature and speech that one sometimes notices in bilin- throws light on the problem. guals." A bilingual person, he suggested, Children who have been deaf from an early may become as competent as a monoglot if age afford one situation in which mental he is prepared to concentrate on one of his development must occur without the usual languages, as for. example Conrad did, until accompaniment of language growth. These that language becomes the natural medium deaf children who have not learned to use of his thought. words show retardation in mental develop- I On the basis of these findings it is suggested ment and even in their performance in non- that restriction of the normal language verbal tests of intelligence. This is the growth of children may depress the standard general finding: of twelve studies from I9I5 of their performance in an intelligence test.f onwards, summarized by Pintner, Eisenson The hypothesis to be tested then is that a and Stanton (94I), all except one found that portion of the observed negative correlation deaf children were on average mentally between family size and intelligence test retarded. A recent investigation by Blair score is due to the limiting influence on Hood (1949), using Alexander's Performance general mental development exerted by an Test as a measure of intelligence, did not sup- environmentally restricted development of port this view; but Watts and Slater (I950) verbal ability. If this hypothesis were justi- have shown that the standardisation of this fied it might afford a clue to a partial resolu- test is misleading. In Aberdeen, on a small tion of the paradoxical results of the Scottish sample, Smith (I952) reported that the Mental Survey, in which a negative correla- mental retardation of deaf children appeared tion was established but no decline in intelli- to become progressively more serious as age gence observed. If part of the negative increased. Pintner and his colleagues also correlation were due to the effect of family summarized five studies of the language environment, then a general reduction" in the development of the deaf, all of which con- size of families would result in an artificial- firmed the very severe retardation suffered rise in mean score which might be sufficient by these children. to mask a real decline : a genetical loss would Other children whose language develop- nevertheless be occurring in proportion to ment is unusual are the bilinguals, whose the extent to which intelligence is inherited average performance in verbal intelligence 'and the extent to which the negative correla- tests is inferior to that of monoglots. Smith tion is due to the limiting of families' by (I923), Saer (I923) and others have found intelligent parents. Other explanations of this result in Wales, though Barke and the situation have, of course, been made- Williams (I933, I938) found no inferiority of for example, the explaining of the observed bilinguals on a non-verbal test of intelligence. rise in score in terms of the test-sophistica- Smith (1948), using a non-verbal test among tion of the I947 children-and such influ- 34 THE'EUGENICSREVIEW ences as this may well be operating to total. Their ages ranged from io2 to IT5, 9omplicate the situation still further. but again more than 90 per cent of them lay Evidence for or against the hypothesis was between iii and I2-3, and the mean age soiught in three ways: firstly; by finding if was II7. The mean family size was 3-38. the method of partial correlation showed a The tests used were Moray House Tests 4I- negative correlation between verbal ability and 42, and Moray House English Test I9. and family size independent of any associa- The possibility of excluding all those whose tion with general intelligence; secondly, by ages lay outside a twelve-month range (to finding if the correlation between test score make the groups more strictly comparable and family size differed from test to test with the eleven-year-old group of the when several tests affected in varying degrees Scottish Mental Survey) was considered; by verbal ability were applied to a single but it was thought inadvisable because of the group of children; and thirdly, by checking fact that test scores were expressed as the prediction *from the hypothesis that, if quotients to eliminate the effect of age, and verbal intelligence tests were applied to because such a course would have excluded children at different ages, the negative cor- the brightest and the dullest sections of the relation betwee' family size and intelligence group instead of making it more representa- test score would be more marked at later tive. ages when the cumulative effect of environ- The tests in this part of the investigation ment began to show itself. were those used by the Education Authority for its transfer procedure, and so were not. Partial Correlation altogether suitable for the present purpose. The first test of the hypothesis, then, in- The tests of intelligence were verbal tests, volved determining whether a negative corre- so that to use these for the holding constant lation between verbal ability and family size of intelligence made likely an under-estimate existed independently of any association of the relation between verbal ability and with intelligence. The method of partial family size. Tests of English attainment correlation was applied to the test scores of were used for the measure of verbal ability; two large groups each of about 2,500 chil- though scores in such tests are corrupted by a dren. These were children at the stage of schooling factor, they do have a considerable transfer from primary to secondary educa- verbal loading. It was clear that a certain tion in Aberdeen. Such transfer groups are negative correlation between the English approximately equivalent to a cross-section score and family size would be attributable of the child population of the city. merely to the dependence of English attain- The first of thesew transfer groups, the I949 ment on intelligence. But if the hypothesis group, numbered 2,709 in all, and for 2,56I were correct, if the largeness of certain of these it was possible to obtain data on size families operated to limit the verbal develop- of family, position in the family, and scores ment of their members, the holding constant in two intelligence tests (Moray House Tests of intelligence test score would still leave a 39 and 40) and a test of English attainment negative partial correlation between family (Moray House English Test i8). This was a size and English score. If, on the other hand, 94 per cent coverage of the group. Their the environmental influence of the size of ages ranged from IO*5 to I2-Io, but more family were negligible, and if the negative than 90 per cent of them lay between the association of intelligence and family size ages of ii-i and 12-3, the mean age being were due only to the limiting of families by II*7. The mean family size was 3.4I. intelligent parents, then one would expect The I950 transfer group numbered 2,638 no association between family size and Eng- in all, and by giving a second opportunity of lish score except such as was due to the testing to those who were absent on the correlation of English score and intelligence. appointed date it was possible to obtain full In that case the partial correlation would be ,information on 2,607, or 99 per cent of the zero. It is maintained in this study that both -FAMILY ENVIRONMENT I35 *causes are at work. If this were so, the result still be negative and significantly different which might be predicted would be that the from zero (due to the first of the causes). partial correlation of family size and English This prediction was confirmed by the analysis score, with intelligence test score held con- of results. stant, would be substantially smaller than The means and standard deviations given the zero-order coefficient (due to the second in Table I show that the groups are repre- of the causes mentioned above), but would sentative samples. TABLE I. MEANS AND STANDARD DEVIATIONS OF QUOTIENTS I949 Transfer Group 1950 Transfer Group 2,56i cases 2,607 cases Test Mean S.D. Test Mean S.D. M.H.T.39 ... IOIP37 I3.8 M.H.T.4I ... I02-59 I3.8 M.H.T.40 ... io6*30 14.8 M.H.T.42 ... io6.3I 15.2 M.H.E.I8 ... I05'33 I5.3 M.H.E.I9 ... I03-35 I5.6 The mean scores in each test for each size correlation a check on linearity of regression of family were calculated, grouping together was made. The partial correlation coeffi- all sizes of eight and above because of the cients are given in Table III. All these smallness of their numbers. The tendency partial correlations are hegative and sig- for mean score to drop with increasing size of nificantly different from zero. They are, family can be seen in the results of all the however, substantially smaller than the tests, as is shown in fig. i. zero-order coefficients. Though this first prediction was confirmed FIG. I. MEAN TEST SCORES AT DIFFERENT SIZES there OF FAMILY are too many sources of ambiguity in the method of partial correlation for it to be Quotient Family Size -- 23 4 5 6 78+ 1 2 3 4 altogether satisfactory; and it was clearly 6 73+ 2 34 5 79+ desirable that alternative methods of Ito ~ *.M.'H.'T M.HE. approach to the problem should be tried to 10 I33 - 39 0 IS see if further support could be obtained for 104 2 3 5 6 the hypothesis. 102- 409 Different Verbal Loadings 96 In the second part of this investigation a 9 233 - i number of tests, depending in varying 32- - 10- .90 - degrees on verbal ability but otherwise as far as possible comparable with each other, 2 34 5 6 7s*+ 2 3 4 56C 78+ 2 3 4 SG 78* were applied to a single group of children. if 2 - M.-IT. M..TMHE The argument at this stage was that if the 190 negative correlation of intelligence test score 305-7919 and family size was seriously affected by the 106 family environment operating on verbal 102 100 4 development in the manner suggested, the negative correlation would probably be 99~ ~ IN9 greater with tests much dependent on verbal 2392 ability, and would tend to be less with tests more nearly independent of verbal ability, provided that the tests were comparable in The figures give the numberILIof cases forjI each other respects. Thus a test of vocabulary LI group of tests should show a relatively large negative The same results are expressed as correla- correlation with family size; a verbal test of tion coefficients in Table II. intelligence should show a certain negative Before applying the method of partial correlation with family size; and a non- 36 THE EUGENICS REVIEW

TABLE II. CORRE3LATION OF FAMILY SIZE AND TEST SCORES Correlation of family size and: Verbal intelligence test M.H.T.39 ...... - -320 I949 ,, ,, ,, M.H.T.40 ...... -335 Transfer M.H.T.39 and M.H.T.40 combined ...... -.333 Group English attainment test M.H.E.i8 ...... -.3I8 N=2,56I Verbal intelligence test M.H.T.4I ...... - 323 I950 ,, ,, ,, M.H.T.42 ...... - .334 Transfer M.H.T.4i and M.H.T.42 combined ...... -.334 Group

English attainment test M.H.E.ig *...... -*342 N=2,607 TABLE III. PARTIAL CORRELATION COEFFICIENTS Partial correlation between family size and: Holding constant: M.H.E.i8 (English) ...... M.H.T.3g (Intelligence) ...... -*o8I ,, ,, ...... M.H.T.40 ... -046 ,, ,, ...... M.H.T.39 and M.H.T.40 ...... -044 Number of cases 2,56I M.H.E.I9 (English) ...... M.H.T.4I (Intelligence) ...... -*13I ,,...i,, ...... M.H.T.42 ,, ...... -*o9I .... M.H.T.4i and M.H.T.42 ...... -*Io9 Number of cases 2,607 verbal test of intelligence, being only slightly ligence quotients were below 50 (Stanford- influenced by family environment, may be Binet), were included in the lowest class expected to show a small negative correlation interval of the distributions of scores, in with family size. If, on the other hand, the accordance with the practice of the Scottish negative correlation is relatively independent Mental Survey. This made the number for of this verbal environmental influence the whom results were available exactly 2oo. variations in the verbal loadings of the dif- The missing six were: three children from ferent tests will not be accompanied by private schools, one deaf child, one invalid, parallel variations in the size of the correla- and one with an irregular school history. tion. Among the twelve measures which were With the limited facilities available, how- obtained for this sample of loo, there were ever, increasing the number of tests neces- four tests of intelligence, two non-verbal- sarily involved reducing the number of Progressive Matrices (I938), untimed, and subjects tested. Consequently it was decided Jenkins' Scale of Non-Verbal Intelligence I to apply as wide a variety of tests as possible -and two verbal-Moray House Tests 4I to a small group which would be a random and 42. The vocabulary items of Moray sample of the population, making sure that House English Test I9 were scored separately the sample was genuinely representative of to give an estimate of range of vocabulary; the population and was fully located and and the remaining measures were of various covered by the testing. aspects of scholastic attainment. The sample selected consisted of all the The scores in the standardized tests were children in Aberdeen who were born in the at first expressed as quotients in accordance month of April I939-a total of 2o6. The with the published norms; but considerable choice of a single month avoided the necessity differences emerged between the mean of age allowances and made possible the use quotients in the four intelligence tests, in of raw scores in calculation. The locating of spite of the fact that the group was a the sample was done through the co-opera- tion of the headmasters, and was checked random sample of the population and all the during personal visits to the schools, against tests were applied in the same order at the Education Authority records and by per- same time, or at nearly the same time. Since sonal letters. correlational analysis of these results was Test results were obtained for I98 of these proposed it seemed desirable to impose new children; and two others, in a special school normal distributions on the raw scores in all for the mentally handicapped, whose intel- the tests, giving each a mean of IOO and a FAMILY ENVIRONMENT 37 standard deviation of 15. Means and stan- tions in the Jenkins Test are printed in small dard deviations are shown in Table IV: type and difficult to interpret. This explana- grouping of scores has introduced small tion is supported by the comparatively high

TABLE IV. MEANS AND STANDARD DEVIATIONS OF QUOTIENTS Old Quotients New Quotients Test Mean S.D. Mean S.D. Progressive Matrices (1938) ...... I02*50 13.0 99-98 I4.6 Jenkins' Non-Verbal Scale I ...... IO7'95 14.3 100-05 I4.6 Moray House Test 4I ...... 103*55 I3.5 99.72 I4.8 Moray House Test 42 ...... io8.30 I5'4 99 90 I4.9 Moray House English Test I9 ...... 104-28 I5.5 99.80 1-4.6 Vocabulary ...... - 99.72 14.6 deviations from the imposed statistics. The correlation between the Jenkins Test and the mean family size was 3-I8 and the standard Vocabulary Test of .73, compared with a deviation of family size was Igo. correlation of *56 between Matrices and Using the new quotients, correlations were Vocabulary. calculated between family size and scores in However, there are at least two alternative these measures, and are given in Table V. explanations of why the correlation involving the Matrices Test should be smaller than the-- TABLE V CORRELATIONS OF FAMILY SIZE AND TEST SCORES others: Correlation of family size and: (i) It might be due to differences in reli- Progressive Matrices (I938) ...-I97 Jenkins' Non-Verbal Scale I ... -*249 ability between the tests : unreliability Moray House Test 4I ...... -*298 of any of the tests would depress the Moray House Test 42 ...... -303 of that test with any other Moray House English Test i9 ... -*289 correlation Vocabulary ...... -286 measure; (ii) It might be due to differences in The main interest in these results lies in validity between the tests : it might be the difference in the correlation between argued that the " intelligence " mea- family size and intelligence test score with a sured by the Matrices Test is a different non-verbal test (Matrices -*I97; Jenkins type of ability from that involved in -249) and that with a verbal test (M.H.T.4I the Moray House Tests,.and that the -298; M.H.T.42 -303). The significance for the differences in the cor- of these differences cannot be calculated relations is that the tests are measuring exactly, for the family size measure is dis- different abilities. crete; but it was tested by Cochran's The first of these possibilities can be tested formula* for a difference between two cor- quite simply by means of correction for relation coefficients based on the same ran- attenuation: such correction had virtually dom sample. The difference between the no effect on the difference between the cor- correlation involving the Matrices Test is relations. The second is rather difficult to significantly different at the -o5 level from test, but a factor analysis of eight of the the correlations involving Moray House measures made possible a comparison of Tests 4I and 42. None of the other differ- the " g " loadings of the various tests. From ences reaches the -o5 significance level. the results there seemed to be no basis For the most part the results are in accord- for the suggestion that the differences be- ance with the prediction that the non-verbal tween the correlations could be attributed tests would show a smaller negative correla- to differences in the " g " loadings of the tion with family size than the verbal tests. tests. The Jenkins Test is intermediate between the There remains the second part of the pre- non-verbal Matrices and the verbal Moray diction, that the vocabulary test would show House Tests, perhaps because the instruc- the highest negative correlation with family * Lindquist, 1940, P. 2I8. size. This was not supported by the results: THE EUGENICS REVIEW all the correlations involving verbal tests, seven, nine and eleven were"'obtained for both verbal intelligence and English attain- I78 of the 200 children in the random sample inent, are of approximately the same magni- described above. The test at age seven was tude. The vocabulary test, however, was Moray House Picture Test; at age nine, shorter than any of the other tests. In case Schonell's Essential Intelligence Test. These it might be thought that the test form of all were compared with the results of Moray these measures was introducing spurious House Test 4I at age eleven. The means and results the correlation was calculated between standard deviations given in Table VI show family size and marks in a composition that the group is representative of the written by these children, and was -253. population. The intercorrelations of the tests were Different Ages between Moray House Picture Test and A third method of testing the hypothesis is Schonell's Test, -738; between Schonell's to find if the age at which the intelligence Test and Moray House Test 4I, .856; and test is administered affects the size of the between Moray House Picture. Test and correlation. A preliminary prediction may Moray House Test 4I, -748. be made that the negative correlation, if it is The correlations of family size and intel- subject to marked environmental influence, ligence test score at the different ages were will tend to be greater at later ages when the at age seven, -*2099; at age nine, -*226; cumulative effect of environment begins to at age eleven, -289. The correlation is show itself. smallest at age seven and increases as the Ideally to test this prediction one should children grow older. Because of the small apply the same test to the same children at number of cases the differences are not different ages. However, at different ages the significant. children would score on different parts of a Significant differences emerged from the test and therefore would in effect be taking testing of large groups at different ages. In different tests, although these might be Aberdeen a representative group of about printed all in a single booklet. Consequently I,200 children enter the primary schools the need to use different tests at different every six months. Moray House Picture ages is a weakness in the design of this part Test was applied to one such group of I,236 of the investigation. The performance of a children at age seven; and Schonell's Essen- small number of children was studied over a tial Intelligence Test to one group of I,270 period of years; and the results from this children at age nine. These two groups were small group were checked against the testing compared with four groups tested at age of large numbers at different ages. eleven plus and described previously as the Scores in verbal intelligence tests at ages I949 and I950 Transfer Groups. This

TABLE VI. MEANS AND STANDARD DEVIATIONS Small group: N Age Mean S.D. Moray House Picture Test ...... 178 7 99-22 14.5 Schonell's Essential Test ...... 9 IOI*I9 i6-8 Moray House Test 4I ...... II I02'79 I3.8 Family size ...... 7 300 1.93 ,, ,, ...... 9 3.15 1.94 ,, ,, ...... II 3.28 1.94 Large age-groups: Moray House Picture Test ...... I,236 7 I04.50 15.5 Schonell's Essential Test ...... 1,270 9 104-63 I7.I Moray House Test 39 ...... 2,56I II IOI37 I3.8 Moray House Test 40 ...... io6:30 14.9 Moray House Test 4I ...... 2360 1028 59 I3.8 Moray House Test 42 ...... 2, 07 IO63I 15.2 Family size ...... I,236 7 2.77 IP54 ,, ,, ...... 1,270 9 3°00 I.70 ,, ,, ...... 2,56I I I 3.41 I.87 ,, ,, ...... 2,607 II 3.38 I.87 FAMILY ENVIRONMENT eleven-year-old group numbered 5,I68 in all, ferences on the family-size side of the 2,56i tested on Moray House Tests 39 and correlation. 40, and 2,607 on Moray House Tests 41 and The result of such a process was to raise 42. The means and standard deviations the correlation at age seven from -*209 to given in Table VI show that all the groups -*225; to raise the correlation at age nine are comparable and representative samples. from -226 to -232; while the correlation In the large age-groups the correlation of at age eleven remained - *289. This brings intelligence test score and family size was the coefficients slightly closer, but the at age seven, - 256; at age nine, -*287; differences remain. and at age eleven, - 333. These coefficients The second alternative explanation, attri- are all larger than the corresponding coeffi- buting the results to differences in test cients in the small group, but they repeat the reliability, may also be tested simply. Cor- same pattern. The negative correlation is rection for attenuation, using the reliability greater when the test is applied at a later age. coefficients given in the test manuals, was The difference between the correlation at age applied to the revised coefficients which have seven and at age eleven is significant at the just been obtained. The result was to raise *oI level. the coefficient at age seven to -230, at age It is difficult to see why the coefficients nine to - *242, and at age eleven to -293. from these large groups should be greater This has very little effect on the differences than the corresponding results from the between the correlations. small group. Vernon (I95I) reported a cor- The question of differences in validity is relation of -34 between intelligence and not so easy to handle. Moray House Test 4I family size in a group of io,ooo recruits aged was selected for this comparison because it i8: however, the tests in that investigation appears to resemble the other tests more were both verbal and non-verbal tests, so closely. It is certainly similar to Schonell's that his results are not directly comparable Essential Test; and it differs from Moray with these. House Picture Test principally in that the The differences between the correlations at latter does not require ability, age seven and at age eleven may be due to although the oral instructions of the Picture any of three causes. They may be due to: Test are just as verbal as those of the other (i) Differences on the size side of the tests. It is clear from results mentioned correlation, because the families are earlier in this paper that if a non-verbal test most nearly complete at age. eleven; had been chosen at age eleven for this com- parison the correlation between test score (ii) Differences on the test side of the and family size would have been much correlation, either (a) differences in smaller. However, it is the verbal element test reliability, or (b) differences in that is of chief interest in this investigation. validity; It remains to be said that since all these (iii) Genuine differences in the degree of calculations were based on the I78 cases of relationship, through the increasing the small group the numbers are too small to effect of environmental influence. yield significant differences at any point. It is only in the small group of I78 cases Significant differences might be expected to that the nature of the data allows a precise emerge from numbers as large as those in the testing of these alternative explanations. age-groups; but since these are separate The first alternative, attributing the result samples of the population it is not possible to differences in the completeness of families, to make an accurate calculation of the effect may be tested easily. The family size at age of incomplete families. A new but rather eleven is the nearest available estimate of the complicated method of estimating the degree completed size, and if the test scores at age of incompleteness of family among an age- seven and nine are measured against family group of children (which will be described size at age eleven, this will eliminate dif- elsewhere) suggests that the effect of incom- 40 THE EUGENICS REVIEW plete families in the large age-groups is influence of the size of family on verbal approximately the same as that found in the development and through it on general small group. If this assumption is valid, and mental development. At the same time, at corrections for incomplete families and un- each stage of the investigation it seemed clear reliability of tests are applied to the cor- that the whole of the negative correlation relation coefficients obtained from the large could not be explained in terms of this age-groups, theresulting corrected coefficients environmental influence. Others who have at age seven and at age eleven show a differ- worked on this problem on a nation-wide ence which is still significant at the *o5 level. scale do not deny a certain amount of The disposal of two of the alternative environmental influence, and would prob- explanations suggests that the third explana- ably wish attention to be drawn to the tion is the correct one, that the differences substantial negative correlation which re- between the negative correlations of intelli- mains when the environmental influence is gence test score and family size at different allowed for. Such remaining negative cor- ages are due to the cumulative effect of relation, it would seem, must have the effect environmental influences. of depressing the trend of national intelli- gence, a tendency which may easily be Summary and Conclusions masked by environmental influences. A hypothesis has been advanced that the environment of the large family-the limited REFERENCES Ballard (I934), Thought and Language. University of amount of contact between parent and child, London Press. and the consequent retardation of the child's Barke (I933), Brit. J. Educ. Psychol., 3, 237-50. to depress the Barke and Williams (I938), Brit. J. Educ. Psychol., 8, verbal development-tended 63-77. environmental component of a child's test Cattell (1950), Eug. Rev., 42, I36-48. score. This was based on previous work on Christophersen (I949), Bilingualism. Methuen. Davis (I937), Inst. Child Welfare Monog. Series, 14. the comparison of orphanage children and University of Minnesota Press. only children, on studies of bilinguals, deaf Day (I932), Child Development, 3, I79-99. children and twins. It was not suggested Dewey (I909), How We Think. Heath. Head (I926), Collected Papers on Aphasia and Kindred that this cause alone operated to produce the Affections. Cambridge University Press. negative correlation of family size and Hood (I949), Brit. J. Educ. Psychol., 19, 210-I9. that con- Lindquist (1940), Statistical Analysis in Educational intelligence test score, but only it Research. Houghton Mifflin. tributes to the correlation to an extent McCarthy (I930), Inst. Child Welfare Monog. Series, 4. sufficient with other influences to mask any University of Minnesota Press. McCarthy (1946), Chapter X, Manual of Child possible downward trend in national intelli- Psychology (editor, Carmichael). Wiley. gence and to prevent the use of test scores of Mehrotra and Maxwell (I949), Population Studies, 3, age-groups of children for predicting future 295-302. Pintner, Eisenson and Stanton (I94I), The Psychology intelligence levels in the nation. of the Handicapped Child. Crofts. Predictions from this hypothesis were Royal Commission on Population (1949), Report. methods: partial correla- Cmd. 7695. H.M.S.O. tested by three by Saer (I923), Brit. J. Psychol., 14, 25-38. tion of family size and verbal ability with Scottish Mental Survey (I949), The Trend of Scottish intelligence held constant; by correlation of Intelligence. University of London Press. Smith, C. (I948), Mental Testing of Hebridean Children family size and several tests with different in Gaelic and English. University of London Press. verbal loadings ; and by correlation of family Smith, D. (I952), Brit. J. Educ. Psychol., 22, 71-2. size and intelligence at different ages. The Smith, F. (I923), Brit. J. Psychol., 13, 27I-82. Terman and Merrill (I937), Measuring Intelligence. fact that the hypothesis has survived these Harrap. three tests does not, of course, mean that it Thomson (1924), Instinct, Intelligence and Character. : are other Allen & Unwin. is correct there still possible Vernon (I95i), Eug. Rev., 43, 125-37. explanations of the results. But it seems that Watts (I944), The Language and Mental Development part of (though not all) the negative correla- of Children. Harrap. Watts and Slater (I950), The Allocation of Primary tion of family size and intelligence test score School Leavers to Courses of Secondary Education. may be attributed to an environmental Newnes. Eugenics Review, Vol. XLV, No. 1