Chapter 13 (p. 267-283) in Learning Potential Assessment, Theoretical, Methodological and Practical Issues. Edited by J.H.M. Hamers, K. Sijtsma & A.J.J.M. Ruijssenaars (1993).
Amsterdam: Swets & Zeitlinger
The Snijders-Oomen nonverbal intelligence tests:
general intelligence tests or tests for learning potential?
P.J. Tellegen & J.A. Laros
Traditional tests for general intelligence (GI-tests) like the Stanford-Binet and the Wechsler intelligence tests are criticized by advocates of Learning Potential tests (LP-tests) on the point that these tests measure the end result of prior learning rather than learning potential. By merely reflecting the end result of prior learning GI-tests would underestimate the learning ability of persons who have had fewer opportunities to acquire the knowledge and skills to perform well in a test situation. In particular, members of ethnic minorities, persons from lower socioeconomic background and persons with learning problems would be at a disadvantage when tested with a GI-test. A related point of criticism implies that GI-tests provide no information on the growth in performance to be expected given optimal learning conditions. As a result these tests would not discriminate sufficiently between mentally retarded and learning disabled children.
Common to LP-tests is the inclusion of training in the test design, either as a separate phase or incorporated in a single test administration. The aim of the training is to eliminate differences due to previous educational or cultural opportunities and to optimize the learning conditions. Although research on learning potential has been going on for several decades (see Guthke, chapter 3 in this volume), practical instruments for general assessment have only recently become available.
GI-tests have also been criticized on the basis of their contents by advocates of culture fair intelligence tests. Because the tests often make an appeal to specific language skills, both in test contents and instructions, these tests would place members of cultural minority groups at a disadvantage. This argument also applies to persons with hearing-, speech- and language problems. For all these groups, low performance on a GI-test might primarily reflect poor verbal knowledge instead of poor reasoning or learning ability. This criticism of GI-tests has led to the development of nonverbal intelligence tests which aim at minimizing the reliance on acquired knowledge and verbal ability, such as Raven's Progressive Matrices (Raven, 1938) and Cattell's Culture Fair Intelligence Test (Cattell, 1950).
In the early forties, Snijders-Oomen (1943) constructed a nonverbal intelligence scale (SON) intended for the assessment of deaf children. Intelligence was defined by her in terms of learning ability; the extent to which children could profit from instruction at school. The SON-test developed by Snijders-Oomen was the first test that covered a wide area of intelligence without being dependent on the use of language. The scale has been revised several times and is especially suited for the intelligence assessment of immigrant children and children with hearing-, speech- and language problems. In the context of learning potential tests a closer look at a nonverbal test like the SON can be valuable in showing that explicit training is not the only alternative to general intelligence tests for 'fair' testing of special groups.
In this chapter the latest revision of the SON-test, the SON-R 5.5-17, will be described. After a short summary of the history of the SON-tests and the characteristics of the SON-R, the most important psychometric qualities and research results with hearing and deaf subjects will be reviewed. In the discussion special attention will be given to the similarities and differences of the SON-R compared to general intelligence tests and learning potential tests.
History of the SON-Tests
In her work with children at an institute for the deaf Snijders-Oomen was confronted with problems of assessing the learning ability of children who were severely handicapped in their language development. General intelligence tests were not suited for this purpose due to reliance on verbal skills, while nonverbal tests at that time consisted mainly of performance tests related to spatial abilities (like mazes, form boards, mosaics). After extensive experimentation with existing and newly developed tasks she constructed a test series which also included nonverbal subtests related to abstract and concrete reasoning. Capacities for abstraction and combination were considered especially important for the ability to participate in the educational system (Snijders-Oomen, 1943, pp. 25-28). Mental age norms were constructed for deaf children from 4 to 14 years of age.
In the subsequent revision the test series was expanded and standardized for deaf and hearing children from 3 to 17 years (Snijders & Snijders-Oomen, 1970). With the second revision, different series of tests were developed for younger and older children respectively, the SON 2.5-7 (Snijders & Snijders-Oomen, 1976), and the SSON for the ages of 7 to 17 years (Starren, 1978). The latest revision, published in 1989, is the SON-R 5.5-17 (Snijders, Tellegen & Laros, 1989; Laros & Tellegen, 1991). A new revision of the test for the preschool age group will be published in 1994. Common to the revisions of the SON-tests is the primary goal to examine a broad spectrum of intelligence without being dependent on language. Due to the nonverbal character of the SON-tests, the test materials can be used internationally without modifications; the manual of the SON-R has been published in the English, German and Dutch languages.
The SON-R 5.5-17
Composition of the SON-R 5.5-17
In sequence of administration the test series consists of the following 7 subtests:
- Categories: The subject is shown three drawings of objects or situations that have something in common. The subject has to discover the concept underlying the three pictures and is required to choose, from five alternatives, those two drawings which depict the same concept. The difficulty of the items is related to the degree of abstraction of the underlying concept. For example, in an easy item the concept is 'fruit' and in one of the most difficult items the concept is 'art'.
- Mosaics: Various mosaic patterns, presented in a booklet, have to be copied by the subject using nine red/white squares. There are six different sorts of squares. With the easy items, only two sorts are used while all six sorts are used with the difficult items.
- Hidden Pictures: A certain search object (for instance a kite) is hidden fifteen times in a drawing. The size and the position of the hidden object varies. After focusing on the search object, the subject has to indicate the places where it is hidden.
- Patterns: In the middle of a repeating pattern of one or two lines a part is left out. The subject has to draw the missing part of the lines in such a way that the pattern is repeated in a consistent way. The difficulty of the items is related to the number of lines, the complexity of the line pattern and the size of the missing part.
- Situations: The subject is shown a picture of a concrete situation in which one or more parts are missing. The sub-ject has to choose the correct parts from a num-ber of alternatives in order to make the situa-tion logically coherent.
- Analogies: The items consist of geometrical figures with the problem format A:B=C:D. The subject is required to discover the principle behind the transformation A:B and apply it to figure C. Figure D is not presented and has to be selected from four alternatives. The difficulty of the items is related to the number and the complexity of the transformations.
- Stories: The subject is shown a number of cards that together form a story. The subject is given the cards in an incorrect sequence and is required to order them in a logical time sequence. The number of cards that are presented varies from four to seven.
The diversity in tasks and testing materials has the advantage of making the test administration attractive for the subjects. Categories, Situations and Analogies are multiple choice tests, the remaining four tests are so called 'action' tests. In the action tests the solution has to be sought in an active manner which makes observation of behaviour possible. Although no observation system is provided with the SON-R, and no data regarding the reliability and validity of observations were collected, many users of the SON-tests appreciate the possibilities for behaviour observation. It is the main reason why the SON-'58 remained in use after the publication of the SSON as in the latter test all subtests were in multiple choice form.
One can divide the SON-R into four types of tests according to their contents: abstract reasoning tests (Categories and Analogies), concrete reasoning tests (Situations and Stories), spatial tests (Mosaics and Patterns) and perceptual tests (Hidden Pictures). The abstract reasoning tests are based on relationships that are not bound by time and place; a principle of order has to be derived from the presented material and applied to new material. For nonverbal testing of abstract reasoning, classification tests and analogy tests are widely used. In the concrete reasoning tests the objective is to bring about a realistic time-space connection between objects. Emphasizing either the spatial dimension or the time dimension leads to two different test types. In the so-called completion tests (Situations), the task is to bring about an imperative simultaneous connection between objects within a spatial whole. In the other type (Stories), the object is to place different scenes of an event in the correct time sequence. The concrete reasoning tests show an affinity to tests for social intelligence in which insight in social relationships and behaviour is emphasized. In the spatial tests a relationship between parts of an abstract figure has to be established. Mosaics is a widely known test-type which was included in the earlier SON-tests; the new subtest Patterns is especially developed for the SON-R. In the perceptual test, Hidden Pictures, one must discover a certain figure hidden in an ambiguous stimulus pattern. This subtest, which is also new for the SON-tests, represents the factor 'flexibility of closure', differentiated by Thurstone.
In contrast to the earlier versions of the SON-tests, the SON-R does not include short-term memory-span tests. As Estes (1982) notes, the way information is organized and retrieved from long-term memory seems much more relevant than short-term memory in assessing the ability of children to succeed in school, where virtually all instruction is presumably intended to deal with long-term memory for the material learned. Although we tried to develop alternatives for the short-term memory tests, it was found to be too complicated and time consuming to integrate nonverbal subtests concerning long-term memory and memory strategies in the SON-R.
Construction of the subtests
The subtests of the SON-R have been systematically constructed on the basis of a theory of item difficulty. The intention of such a theory is to cover the most important factors that contribute to the difficulty of items in a subtest. With the help of such a theory items can be ordered as subsequent, logical steps in the mastery of a specific problem type. A theory which is successful in explaining the progressive difficulty in a subtest has two important advantages. In the first place, it creates the possibility of designing items with a certain degree of difficulty, and of performing a systematic test construction. Secondly, one obtains a rational basis for interpreting failure at a certain level of difficulty. Especially for the subtests Mosaics, Patterns and Analogies we succeeded in developing effective theories of item difficulty. The unidimensional scaling model developed by Mokken (1971) was used in selecting the items. In terms of the model, the subtests are reasonable strong scales with H-coefficients close to .50.
Like most intelligence tests for children, the SON-R is individually administered. Group-administration is less suited for nonverbal instructions and for motivating young subjects, and would exclude behaviour observation. The role of time scoring is kept to a minimum. In this sense, the SON-R is a typical power test; there is a large variation in the difficulty of the items, while there is sufficient time for solving each item. The time needed to administer the SON-R varies from 1 to 2 hours with an average of 1.5 hours. There is a shortened version of the SON-R, consisting of four subtests: Categories, Mosaics, Situations and Analogies. The administration of this shortened version takes about three quarters of an hour.
For the subtests of the SON-R there are verbal and nonverbal instructions which have been made as equivalent as possible. Nonverbal instruction forms the point of departure, verbal parts are added as accompaniment and not as supplementary information. The two sets are not intended for use as two exclusive alternatives but they give, in a different form, essentially the same information. With deaf and hearing disabled children one can often use an intermediate form by combining the nonverbal instructions with (parts of) the verbal instructions. In practice, the choice between the two procedures is generally not a problem; one adjusts to the form of communication the subject is used to.
In two important aspects the SON-R distinguishes itself from traditional intelligence tests with regard to the test procedure: firstly, by giving feedback to the subject and, secondly, by the use of an adaptive procedure in presenting test items. It is tradition in intelligence testing not to give feedback whether the subject's answer is right or wrong. This tradition is broken in the SON-R because we think that such behaviour is not natural. When no reaction is allowed following an answer, the examiner's attitude can be interpreted by a subject as indifference or, erroneously, as an indication that the answer was correct. In the SON-R, the subject is told whether the answer was correct or incorrect following each item. However, this does not include an explanation of why an answer is incorrect. One of the advantages of giving feedback is that the subject has the opportunity to change his problem solving strategy. Also, when a subject has interpreted the instructions incorrectly, feedback offers the opportunity to adjust.
The second important difference of the test procedure of the SON-R with common test procedure concerns adaptive testing. In intelligence tests for children with a wide age range, the difficulty of the test items has to be very divergent. Presentation of all items to every subject is troublesome for a number of reasons. In the first place, this would greatly extend the duration of the test. In the second place, it is frustrating for young or less intelligent subjects to be required to solve many items that are too difficult, while the motivation of older and more intelligent subjects is reduced when they are required to solve many problems that are too easy. A practical solution, often followed, consists of presenting all items in order of difficulty and applying a discontinuation rule. However, this procedure does not result in eliminating items that are too easy for a specific subject, and the procedure has the effect that the items on which the subject fails often occur in successive order, which can be highly frustrating. In recent years, adaptive test procedures have been developed which restrict the presentation to those items that are most suited for the specific subject. These adaptive procedures have the goal of effectively limiting the number of items to be administered with relatively little loss of reliability (Weiss, 1982). With computerized testing, these procedures can easily be implemented; with non-computerized testing, there are great practical difficulties for the examiner, both in selection and presentation of the most informative items. The SON-R uses an effective adaptive test procedure by dividing the subtests into either two or three parallel series of about 10 items. The difficulty increases relatively fast in the series. The first series of items serves to estimate the subject's general level of performance. The series is broken off after two errors. Those items in the following series that can most effectively improve and refine the measurement are administered by skipping easy items and by stopping again after two errors. This way, the administration is determined by the subject's individual performance and the presentation is limited to the most relevant items. For the examiner this method has the advantage of presenting the items within a series in a fixed sequence. Thus, searching in the test booklet for the item which has to be presented next, takes place only at the beginning of a new series. For the subject, it is motivating that relatively easy items are presented after two errors.
The standardization of the test scores of the SON-R is based on a nationwide sample of 1350 subjects varying in age from 6 to 14 years. Per age group the sample consisted of 150 subjects and was stratified according to sex, educational type and demographic variables. The population was restricted to persons residing in The Netherlands for at least one year who were not suffering from severe physical or mental handicaps.
From 6 to 14 years, test performance strongly increases with age; 66% of the variance of the raw total score is explained by age. To make comparisons between subjects of different ages possible, a standardization of test scores dependent on age is required. In practice, such standardizations are often performed on the separate age samples. For the SON-R a model has been developed in which the cumulative proportions of the raw scores in the nine age groups are simultaneously fitted as a higher order function of raw score and age. This method yields population estimates of the score distributions which are more reliable (by combining the information of all the age groups), more consistent (by imposing constraints on the form of the functions), and which can be computed for any specific age. By using this model it was possible to extrapolate the age norms to 5;6 and to 17;0 years.
Dependent on the age of the subject, the raw subtest scores are normalized and standardized, thus reflecting the relative position of an individual compared to persons of the same age. The total score on the test is based on the sum of the standardized subtest scores. In the SON-R manual, norm tables for 38 age groups are presented. Even more accurate norms are obtained by using the computer program which is supplied with the test. The program computes norms based on the exact age of the subject.
Reliability and generalizability
The reliability of the standardized subtest scores depends on the correlations between the item responses. Since, with the adaptive procedure, almost 50% of the items are not actually administered, the correlations between the item scores are systematically and artificially enhanced because not administered easy items get a score of '1' and not administered difficult items get a score '0'. As a result, usual formulas will overestimate the reliability. A separate study has been conducted to achieve unbiased estimates. In this study the subtests were (almost) completely administered. Score patterns of the complete administration were compared to score patterns computed as if the adaptive procedure had been applied. It appeared that as a result of the adaptive procedure the correlations between the subtests decreased, while the computed alpha coefficient was higher compared to the alpha of the complete administration. The outcome indicates that (averaged over subtests and age groups) the actual reliability with the adaptive procedure is .10 lower than computed by coefficient alpha. The actual reliability of the adaptive procedure is .05 lower than the reliability with complete subtest administration. The corrected reliability of the subtests of the SON-R is .76 on the average. The most reliable subtests are Mosaics, Patterns and Analogies.
In classical test theory, reliability refers to the stability of hypothetic independent repeated measurements (see Lord & Novick, 1968). In the theory of generalizability the items are considered to be a sample from a domain of comparable items and the internal consistency of the item scores indicates how valid it is to generalize from the outcome of the sample to the entire item domain (Cronbach, Rajaratnam & Glaser, 1963; Nunnally, 1978). For homogeneous item sets, both approaches are almost equivalent. For the total score on an intelligence test that is composed of several subtests, all partly measuring separate components, an important distinction between reliability and generalizability can be made. With the reliability of the total score (stratified alpha; Nunnally, 1978, p. 246), the possibilities for generalization remain restricted to the specific contents of the subtests. For the interpretation of individual outcomes it will be more relevant to generalize to the entire domain of comparable subtests, and to consider the subtests as a restricted sample of the domain that is important for the assessment of intelligence. In the latter case, the number of subtests and the mean correlation between the subtests determine the coefficient of generalizability. This can be computed by the usual coefficient alpha in which the subtests are the unit of analysis (Nunnally, 1978, p. 212).
For the SON-R, the reliability of the total score (alpha stratified) is .93. The generalizability of the total score (alpha) increases from .81 at six years to .88 at fourteen years, with a mean value of .85. For the shortened version of the SON-R, the reliability has a value of .90 and the generalizability has a mean value of .77.
Stability through time
Test-retest research has not yet been carried out with the SON-R. In the research with the SON-R with deaf subjects, test results on earlier versions of the SON were available for 434 subjects. The mean correlation of the SON-R with earlier versions of the SON is .76, and is related to the age at administration of the first test and to the lapse of time between the two administrations. As is the case in American research on general intelligence tests, stability increases with age and with shorter time intervals (Bayley, 1949).
The correlations between the standardized subtest scores steadily increase with age. The mean value is .38 at six years and .51 at fourteen years. Correcting for the unreliability of the test outcomes, the correlations increase from .52 to .68 with a mean value of .61. Although the test scores cohere to an important degree, multiple correlations per subtest with the six other subtests show that a substantial part of the reliable variance per subtest is unique and cannot be explained by the other subtests (averaged over the subtests, this percentage is 47% at six years, and 32% at 14 years).
To investigate whether the interrelations can be explained by uncorrelated components, and whether these components correspond to the division of the subtests in tests for concrete reasoning, tests for abstract reasoning, spatial tests and perceptual tests, principal components analysis has been performed on the correlation matrix after correction for attenuation. The dominance of the first component is quite strong; it is the only component with an eigenvalue greater than one and the percentage of explained true score variance is 59% at 6 years and 72% at 14 years. This indicates that the subdivision of the subtests in four categories is not of major importance. For six year olds, the loadings on the first four varimax rotated components confirm to a great extent the above mentioned categorization of the subtests; for the older subjects, most subtests have high loadings on several components. The structural characteristics of different groups (hearing/deaf, native/immigrant) are highly similar.
Presentation and interpretation of the results
Given the above-mentioned characteristics, the total score of the SON-R provides a reliable and generalizable indication of nonverbal intelligence. The subtest scores add information concerning specific abilities. For the interpretation of the standardized subtest scores and the IQ scores, reliability is taken into account in two different ways, namely by representing the scores as norm scores and as latent scores. For both types of scores, the basis of the standardization is that the distribution of true scores has a population mean of 100 and a standard deviation of 15. The norm score and the latent score are different approaches of estimating the true score and they are used for different purposes.
The norm score is defined as the sum of the standardized true score and the error of measurement. This unbiased estimate of the true score is used for hypothesis testing, research on groups, and for computation of the total test score. Although the more common standard scores (with standardized observed scores instead of standardized true scores) are also unbiased estimates of their true scores, these true scores do not have a fixed distribution (the standard deviation is dependent on the reliability) which means that for standard scores on different (sub)tests there is no sensible basis for the comparison of their true scores.
The latent score is the estimate of the true score computed by means of linear regression. In combination with the accompanying probability interval of the true score, the latent score is best suited for individual interpretation of the test results and for intra-individual comparison of subtest scores. The latent scores are presented graphically on the scoring form. In the computation of the latent subtest scores, the correlations between the subtests are used to improve the prediction of the true score of each subtest. To predict the true score of a specific subtest, also the performance on the other subtests enters the multiple regression equation. The problem of exaggeration of intra-individual differences in the profile is thereby avoided.
Latent scores for the IQ are computed in two ways, denoted as specific IQ and as generalized IQ. The regression of the specific IQ is based on the reliability of the total score; the regression of the generalized IQ is based on the coefficient of generalizability. The latter score is the estimated performance on the entire domain of comparable intelligence tests, and is best suited for interpretation of the test result as level of intelligence.
Next to these scores which take reliability into account, some descriptive characteristics of the test results are also presented. The reference age is given for the subtests and for the total score; it represents the age at which a specific test result corresponds with a standardized score of 100. This 'mental' age makes interpretation from a developmental viewpoint possible. The total score is also presented as a standard IQ (fixed population standard deviation of 15) with the corresponding percentile scores for the general hearing population and for the population of the deaf. In contrast to earlier versions of the SON, no separate subtest-norms are computed for the deaf.
Compared to intelligence tests that only present standard scores (which do not take measurement errors, and errors of generalization into account) the SON-R offers several extra possibilities for a psychometrically sound interpretation of test results. These possibilities require additional work when using the norm tables, but when using the computer program all results are automatically computed and printed.
In the standardization research with the hearing subjects, data have been collected to substantiate the validity of the test. Separate research has been performed with deaf children to develop supplementary test norms and to further validate the test for this particular group. The main findings will be summarized below.
Between boys and girls, there is no difference in mean IQ scores. A significant relation (p<.01) with sex is found only for Mosaics; girls score somewhat lower than boys on this subtest.
For hearing as well for deaf native Dutch subjects, there is a relatively strong association between occupational level of the parents and the IQ scores. The mean difference between children of unskilled workers and professionals (the two extremes for six categories of socioeconomic status) is about 15 IQ points. Of the 7 subtests, Analogies shows the strongest relation with occupational level.
In the research with hearing subjects, substantial differences in test performance on the SON-R exist between immigrant children (based on country of origin of the parents) and native Dutch children. The mean IQ score for the Moroccan and Turkish children is 84, compared with a mean score of 100.5 for the native Dutch children. The lag of the other immigrant children is small (mean IQ is 99). Comparable differences occur in the deaf research group, except that the lag of deaf children from Surinam and The Netherlands Antilles is also considerable. For deaf and hearing subjects, ethnic differences in performance concern all subtests, but are most pronounced for Mosaics and Analogies. Neither for the hearing, nor for the deaf immigrant children a relation exists between number of years residing in The Netherlands and the test scores. This indicates that lack of knowledge of the Dutch language is not an important cause of their lower results. The differences between native and immigrant children can for a great part be explained by differences in socioeconomic status of the parents, as most parents of immigrant children belong to the lower occupational levels. The difference between native and immigrant children decreases with about one third after controlling for socioeconomic status.
Because school achievement is strongly related to intelligence, and prediction of school success is an important goal of intelligence assessment, the relationship with school career is one of the most direct indications of the validity of an intelligence test. For the SON-R, the relationship of test performance with school career has been examined by stepwise multiple regression for three indicators which appear to play a different role at different ages. These indicators are differentiation to type of school (like special education, general education), grade repetition, and report marks.
In primary education, the relation of school type with the IQ scores is limited; the difference between pupils of special education and general education is considerable (16 IQ points), but relatively few pupils are in special schools. Grade repetition relates strongly to the IQ scores. A relatively large group of pupils in primary education have repeated one or more grades and they have a lag in IQ scores of almost 19 points. Report marks also add to the explained variance of the IQ scores; for the younger group of primary education this is 10% and for the older group it is 16%. The correlations of the IQ score with school subjects like language, arithmetic, and history/geography are of the same order. The multiple correlation of the different indicators of the school career with the IQ scores is .54 in the age group of 7-9 years, and .60 in the age group of 10-11 years. For the children in secondary education, the multiple correlation increases to .63. For these children the relation is almost completely determined by the differentiation into school type; grade repetition and report marks add little to the explained variance.
In many primary schools a school achievement test is administered at the end of the sixth grade. Scores on this test were available for 49 subjects. The correlation with the SON-R IQ is .66. The correlations with the different parts of the achievement test (language, arithmetic and information processing) are largely similar.
Other intelligence tests
A group of 36 children from an outdoor psychiatric clinic has been tested with the SON-R, the WISC-R (Vander Steene et al, 1991) and the Raven-SPM. The distributions of IQ scores on the SON-R (m=97.1; sd=16.4) and the WISC-R (m=96.1; sd=16.0) are highly similar (Nieuwenhuys, 1991). The scores on the Raven (based on English norms) have the same mean but a smaller standard deviation. The correlation of the SON-R with the WISC-R is .80; the correlation with the verbal part of the WISC-R is .65 and the correlation with the performance part is .79. Both SON-R and WISC-R correlate .71 with the Raven.
Performance of deaf children
Starting with the first version, the deaf population has received special attention in the SON-tests. Next to the nationwide sample of hearing subjects, almost the complete population of deaf pupils from 6-14 years of the Institutes for the Deaf and the Schools for the Partially Hearing, with a hearing loss of at least 90 dB, have been examined with the SON-R.
The total group of 768 deaf children has a mean IQ of 90. The difference with hearing children is reduced to 8.5 points when we control for the proportion of immigrants, which is four times as large as in the hearing group. After controlling for occupational level, the difference between the native deaf and hearing subjects becomes 7.7 points. Further analysis shows that this lag in performance of the deaf children is strongly related to the presence of multiple handicapped children in the deaf population (about 25%). Several causes of deafness, such as complications during pregnancy and birth, and meningitis and encephalitis, can also be the cause of mental retardation. Excluding the multiple handicapped, the lag of the deaf children is 4 IQ points which is mainly related to the subtests for abstract reasoning.
The correlation of multiple handicaps and teacher's evaluation of intellectual insight with the IQ scores is .63 and this increases to .66 by also including specific evaluations of cognitive handicaps, communicative handicaps and accuracy. The IQ scores correlate .49 with the STADO-R, a written language test for the deaf (de Haan & Tellegen, 1986). This test consists of four parts, that is synonyms, word order, idiom, and prepositions-conjunctions.
The SON-tests have been developed as an alternative to general intelligence tests for the assessment of cognitive functioning of various groups of children who are handicapped in the area of verbal communication. With the latest revision, the SON-R, this has resulted in a test series which deviates from general intelligence tests in contents and in administrative procedures. In this section we will compare the SON-R both with GI-tests and with LP-tests.
The main difference between GI-tests and LP-tests is the help which is offered to the subject. In GI-tests items are presented only once, often with minimal instruction, and no training and feedback are given during test administration. With LP-tests help is given in the form of extended instructions, feedback, and training at the level at which the subject fails to succeed. The score on the LP-test reflects test performance as a result of the interactive help procedure.
Although no formal training is given in the SON-R there are several elements of the administration that facilitate learning opportunities during testing. These elements are: (a) the several examples given with each subtest, (b) the feedback which informs the subject whether the answer is correct and (c) the adaptive procedure by which easier items are presented after some failures. In this respect the SON-R shares important aspects of the testing procedure with LP-tests. The element of training is even more pronounced in the SON-test for preschool children. In the SON 2_-7 extensive feedback is given to the child after each failure by presenting the correct solution.
A second consideration for the comparison of tests relates to test contents and the specific abilities that are measured. Most general intelligence tests consist of a verbal and a performance scale. The verbal part, which also includes quantitative reasoning tasks, emphasizes crystallized abilities which are greatly influenced by schooling and also by more general experiences outside of school (Thorndike, Hagen & Sattler, 1986, p. 4). The performance part is more related to spatial-visualization abilities. Subtests that focus on fluid-reasoning abilities, like analogies, classification and series completion are included in either the verbal or the performance part, depending on whether the elements of the items are verbal or figural. The SON-R only contains subtests with a nonverbal content thereby excluding subtests specifically aimed at measuring verbal ability and quantitative reasoning. However, the composition of the SON-R in terms of intelligence factors is wider than the performance part of most GI-tests since it is less dominated by spatial tests. Four of the seven SON-R subtests are fluid-reasoning tests in a nonverbal form.
Research with learning potential has for a large part been carried out with existing verbal and nonverbal subtests. For instance Schroots (1979) used subtests from the Leiden Diagnostic Test (Schroots & Alphen de Veer, 1976); Hamers and Ruijssenaars (1984) used four subtests from several intelligence tests; Spelberg (1987) used subtests from the SON-R for a testing the limit procedure and Resing (1990) used two subtests from the RAKIT (Bleichrodt, Drenth, Zaal & Resing, 1984). LP-tests are not characterized by the specific contents of the tests but by the inclusion of training in the testing procedure. When the goal of these tests is to eliminate differences due to prior opportunities, nonverbal tests and training procedures which do not require verbal skills seem to be appropriate. Such a nonverbal LP-test for general use, the Learning potential test for Ethnic Minorities (LEM), has recently been published in The Netherlands (Hamers, Hessels & van Luit, 1991). In table 1 we have summarized the main differences between the SON-R, the WISC-R, as an example of a GI-test, and the LEM, as a somewhat special example of a LP-test.
Table 1: Comparison of the WISC-R, the SON-R and the LEM
||verbal and nonverbal
||verbal and nonverbal
As this table suggests there is a greater correspondence between the SON-R and the LEM than between the SON-R and the WISC-R. To the extent that possibilities for learning during test administration are offered: the SON-R has a position in between the two other tests. With regard to test contents: the SON-R is more similar to the LEM. Although we classified the LEM as nonverbal, two subtests are related to verbal ability, but they do not make use of meaningful words. One subtest measures the learning of relations between meaningless words and objects and the other measures memory of series of syllables. However, they are nonverbal in the sense that they are not dependent on knowledge of a specific language.
The analysis, thus far, of the different tests leads to the conclusion that the question 'SON-R, a general intelligence test or a test for learning potential?' is too simplistic; for a classification of tests more dimensions are needed. One dimension of ordering tests concerns the possibilities for learning during administration. On this dimension LP-tests score high although there is a great diversity in the amount of help and the type of training that is being offered. Traditional intelligence tests score low on this dimension and the position of the SON-R is somewhere in between. A second dimension concerns the use of a specific language in instructions and test materials. Nonverbal tests like the SON-R, LEM and the Raven aim at minimizing this aspect. A third, and very complex, dimension concerns the different cognitive aspects that are represented by the test, like verbal-, spatial- and reasoning abilities and memory, and the extent to which the measurement of these abilities depends on knowledge learned at school and/or the cultural environment. Not only between, but also within the domains of nonverbal tests, GI-tests and LP-tests, there are great variations in test composition. However, nonverbal tests are more restricted since they do not directly measure crystallized verbal ability.
The differentiation between intelligence tests is also reflected in definitions of intelligence. The aspects of knowledge, problem solving and ability to learn are stressed to different degrees, both in definitions and in tests. Which test is 'the best' can only be determined for specific situations on an empirical basis by looking at the validity with regard to relevant theoretical and practical questions. However, the comparison of tests is a very complex matter, not only between separate studies because of differences in populations and criterium measures, but also within a study it can be difficult to differentiate between the effects of reliability and the multiple factors related to content and administration on the test scores. When, for example, immigrant children score higher on test A than on test B this might be the result of differences in reliability of the tests (when standard scores are used) and not result from differences in contents and procedures.
In our opinion, the research results with the SON-R indicate that the test is a useful instrument for the nonverbal examination of children's intelligence, with high reliability and ample indications of the validity. The variety of tasks and test materials is stimulating for the subject and the adaptive procedure avoids repeated presentation of excessively difficult items. An objection to a nonverbal test like the SON-R might be that the concept of intelligence is substantially narrowed by the exclusion of verbal ability tests. However by including tests for concrete and abstract reasoning - areas that often have a verbal form in general intelligence tests - the contents of the SON-R are not limited to typical performance tests. Although the test can be administered without using language, this does not exclude the importance of verbal abilities for the evaluation of intelligence with the SON-R, as is illustrated by the correlations of the test with report marks and tests for language skills. Verbal intelligence tests often require specific knowledge learned in school. When the main object of using a test is to make predictions concerning school achievement, the absence of verbal tests in the SON-R might reduce its predictive power. If, however, the goal of intelligence assessment is to distinguish between possible causes of poor school performance, a test that is not dependent on specific knowledge is more appropriate. In such cases use of the SON-R is not only indicated for special groups such as deaf and immigrant children, but also suited for children with no specific problems in the areas of language and communication.
Bayley, N. (1949). Consistency and variability in the growth of intelligence from birth to eighteen years. Journal of Genetic Psychology, 75, 165-196.
Bleichrodt, N., Drenth, P.J.D., Zaal, J.N. & Resing, W.C.M. (1984). Revisie Amsterdamse Kinder Intelligentie Test, Handleiding [Revision Amsterdam Child Intelligence Test, Manual]. Lisse: Swets & Zeitlinger.
Cattell, R.B. (1950). Handbook for the individual of group Culture Fair Intelligence Test. Scale I. Champaign, Ill: I.P.A.T.
Cronbach, L.J. Rajaratnam, N. & Gleser G.C. (1963). Theory of generalizability: a liberalization of reliability theory. British Journal of Statistical Psychology, 16, 137-163.
Estes, W.K. (1982). Learning, memory and intelligence. In R.J. Sternberg (Ed.), Handbook of human Intelligence. Cambridge: Cambridge University Press.
Haan, N. de & Tellegen, P.J. (1986). De herziening van een schriftelijke taaltest voor doven [The revision of a written language test for the deaf]. Groningen: Internal report, Department of Personality Psychology, HB-86-828-SW.
Hamers, J.H.M., Hessels, M.G.P. & Luit, J.E.H. van (1991). Leertest voor Etnische Minderheden, Handleiding [Learning test for Ethnic Minorities, Manual]. Lisse: Swets & Zeitlinger.
Hamers, J.H.M. & Ruijssenaars, A.J.J.M. (1984). Leergeschiktheid en Leertests [Learning potential and learning potential tests]. Lisse: Swets & Zeitlinger (2nd edition 1986).
Laros, J.A. & Tellegen, P.J. (1991). Construction and validation of the SON-R 5.5-17, the Snijders-Oomen non-verbal intelligence test. Groningen: Wolters-Noordhoff.
Lord, F.M. & Novick, M.R. (1968). Statistical theories of mental test scores. Reading, Mass: Addison-Wesley.
Mokken, R.J. (1971). A theory and procedure for scale-analysis. The Hague: Mouton.
Nieuwenhuys, M. (1991). Een vergelijkingsonderzoek SON-R, WISC-R en Raven-SPM [A comparative study of SON-R, WISC-R and Raven-SPM]. Amsterdam: Internal report, Department of Developmental Psychology.
Nunnally, J.C. (1978). Psychometric Theory. New York: McGraw-Hill (2nd edition).
Raven, J.C. (1938). Progressive Matrices: A perceptual test of intelligen-ce, 1938, individual form. London: Lewis.
Resing, W.C.M. (1990). Intelligentie en leerpotentieel [Intelligence and learning potential]. Lisse: Swets & Zeitlinger.
Schroots, J.J.F. & Alphen de Veer, R.J. van (1976). Leidse Diagnostische Test: Handleiding [Leiden Diagnostic Test: Manual]. Lisse: Swets & Zeitlinger.
Schroots, J.J.F. (1979). Cognitieve ontwikkeling, leervermogen en schoolprestaties [Cognitive development, learning ability and school achievements]. Lisse: Swets & Zeitlinger.
Snijders-Oomen, N. (1943). Intelligentieonderzoek van doofstomme kinderen [The examination of intelligence with deaf-mute children]. Nijmegen: Berkhout.
Snijders, J.Th. & Snijders-Oomen (1970). Snijders-Oomen Non-verbal Intelligen-ce Scale: SON-'58. Groningen: Wolters-Noordhoff.
Snijders, J.Th. & Snijders-Oomen, N. (1976). Snijders-Oomen Non-verbal Intelligence Scale SON 2.5-7. Groningen: Wolters-Noordhoff.
Snijders, J.Th., Tellegen, P.J. & Laros, J.A. (1989). Snijders-Oomen Non-verbal intelligence test: SON-R 5.5-17. Manual and research report. Groningen: Wolters-Noordhoff.
Spelberg, H.C. (1987). Grenzentesten [Testing the limits]. Groningen: Stichting Kinderstudies.
Starren, J. (1978). De ontwikkeling van een nieuwe versie van de SON voor 7-17 jarigen. Verantwoording en handleiding [The development of a new version of the SON for 7-17 year olds. Manual and Research Report]. Groningen: Wolters-Noordhoff.
Thorndike, R.L., Hagen, E.P. & Sattler J.M. (1986). The Stanford-Binet intelligence scale: Fourth edition technical manual. Chicago: The Riverside Publishing Company.
Vander Steene, G., Haassen, P.P. van, Bruyn, E.E.J. de, Coetsier, P., Pijl, Y.J., Poortinga, Y.H., Spelberg, H.C. & Stinissen, J. (1991). WISC-R, Nederlandstalige uitgave: Verantwoording. [WISC-R, Dutch language edition: Research Report]. Lisse: Swets & Zeitlinger.
Wechsler, D. (1974). Wechsler Intelligence Scale For Children - Revised. New York: The Psychological Corporation.
Weiss, D.J. (1982). Improving measurement quality and efficiency with adaptive testing. Applied Psychological Measurement, 6, 473-492.