Perceptual and Motor Skills, 1977, 44, 767-776.
Perceptual and Motor Skills 1977
BACKGROUND INTERFERENCE PROCEDURE AND DISCRIMINANT FUNCTION ANALYSIS I N PREDICTING CLINICALLY DETERMINED CATEGORIES OF LEARNING DISABILITY1 BARRY L. MALLINGERa Radford CoZlege Summary.-There is a need to determine the extent to which the "Background Interference Procedure" as an adjunct to the Bender-Gestalt can account for criterion variance begond that level predicted by an optimal battery. Discriminant functions empirically dassified subjects into clinical categories of learning disability. A reduced battery of intellective and visual-motor predictors generated two significant functions, accounting for 91 % of the variance. The first dimension reflected over-all intellectual functioning, the second, psychomotor skills. Empirical classification accurately categorized 71 % of all s u b jects across five criterion groups. The funcrions efficiently separated the criteria, but the six Background Interference Procedure predictor variables did not improve prediction. Implications include using the Background Interference Pr* cedure for early screening of learning disabilities and employing discriminant functions for data reduction and construct validation of teachers' and judges' ratings.
The literature indicates there have been numerous problems in diagnosing the group of children with average or near-average intelligence and with learning disabilities, language, and/or perceptual-motor deficits. While definitive diagnosis of learning disabilities is hampered by elusive criteria and by the lack of homogeneity among elements within the syndrome, many experts continue to stress the importance of careful and proper diagnostic procedures (Oettinger, 1971). Two very frequently used psychological instruments for differential diagnosis with young children are the Wechsler Intelligence Scale for Children (Wechsler, 1949) and the Bender Visual-motor Gestalt Test (Bender, 1938). While many researchers and all practitioners suggest careful scrutiny of each child suspected of learning disabilities, dinical and school psychologists can verify that such a procedure is often time-consuming and expensive. It seems obvious that the use of computers could greatly facilitate this procedure. An ideal yet feasible approach would be to combine a thorough diagnostic evaluation of the child with computerized and statistical handling of the extended data. Numerous researchers have demonstrated that clinical prediction seldom betters statistical prediction for groups (Sawyer, 1966). In recent years many re'The author wishes to acknowledge, with gratitude, Dr. Steven V. Owen, University of Connecticut, for his assistance in data collection and analysis. Appreciation is also extended to the Editors of Perceptual and Mot07 Skills, for theic helpful comments in preparing this article. 'Paper presented at the annual meeting of the Virginia Academy of Science, Madison College, Harrisonburg, Virginia, May, 1975. Address correspondence to Barry L. Mallinger, Department of Psychology, Radford College, Radford, Virginia 24142.
768
B. L. MALLINGER
searchers have called for the application of multivariate procedures to dassification and prediction problems in psychology and education (Tatsuoka, 1971). A major restriction of studies investigating classification problems is that many researchers continue to employ univariate techniques, such as simple correlations between a set of predictors and a criterion, in spite of the availability of multivariate procedures. The primary advantage of these latter techniques is that one may use a combination of predictors simultaneously and eliminate predictors which do not account for a significant percentage of unique criterion variance. Several multivariate techniques have been applied in studying the classification of learning disabled school children, but discriminant function analysis is probably the most efficient for maximalIy discriminating among diagnostic groups. Ackerman, Peters, and Dykman (1971) found discriminant functions could separate learning disabled from non-disabled children on the basis of WISC predictor variables. Eaves, Kendall, and Crichton (1972) predicted learning disabilities in kindergarten children using discriminant functions which were generated from a battery of intellective and visual-motor variables. Mallinger, Owen, McCook, and Gable (1973) demonstrated the feasibility and efficiency of using behavioral predictors in a multivariate format to separate various groups of learning disabled children. While the Bender-Gestalt is popular with psychologists (Schulberg & Tolor, 1961) and has been used in a multivariate format for differential diagnosis (Burgess, Kodanat, Ziegler, & Greenburg, 1970), the test is not consistently sensitive in screening deficits in psychomotor performance of learning disabled children. A new procedure, developed to deal with problems often encountered in using the Bender for differential diagnosis, is Canter's (1966, 1968) "Background Interference Procedure." This test has been demonstrated to be more sensitive to visual-motor problems than is the standard Bender. This technique is applicable with school-age children (Adams & Canter, 1969; Kenny, 1971) and is conducive to multivariate analysis (Adams, 1970). The major objective of this research, then, was to determine whether a set of predictor variables, derived from the Background Interference Procedure for the Bender-Gestalt, could account for criterion variance beyond that level predicted by an established battery of visual-motor and intellective variables. METHOD Of the many statistical procedures considered "multivariate," multiple discriminant analysis was used in this research (Finn, 1968). This procedure basically concerns the derivation of dimensions which separate criterion groups. This technique permits the reduction of many predictors to a single weighted composite score (or more than one weighted coefficient in the case of multiple discriminant analysis) which can discriminate among criterion groups. The
PREDICTING LEARNING DISABILITY
769
researcher derives canonical variates (dimensions or functions) which maximize differences between groups on the basis of differences within groups. On the basis of standardized weights, an established battery of predictors was reduced to an optimal set, and significant discriminant functions were generated. Variables having the largest weights were applied to the sample in order to contrast the clinical, a piori diagnoses of judges with the empirical classifications. Proportions of hits and misses in the categorization of the criterion groups were calculated (Dixon, 1968), and discriminant function centroids were then plotted to interpret the dimensions in a meaningful manner.
S~lbjecls The sample consisted of 100 elementary school children, 6 to 7 yr. old, in a suburban Northeastern community. All subjects had been nominated by their teachers to attend classes for the learning disabled, and each child was given an individual psychological evaluacion. The evaluation consisted of the Wechsler Intelligence Scale for Children, the standard Bender-Gestalt, and the Background Interference Procedure. Each child was asked to draw the Bender designs rwice, once under standard conditions and once on special paper. One-half of the children drew the Bender figures on the special paper for the Background Interference Procedure before the WISC was administered, that is, at the very start of the testing. These children received the standard Bender Following the WISC. The order was reversed for the ocher one-half of the sample. This procedure was followed to investigate whether order-of-presentation had any effect on the scores and to control for a possible practice effect. Directions given to the children for the second Bender, whether Background Interference Procedure or standard, were simply: "Remember those drawings you did before! Now, I would like for you to draw the same designs again, just like you did before." When questions regarding the special paper were raised, the examiners answered in a non-committal manner, e.g., "It's up co you," or "Just do the best you can." Each psychologist scored his own tests, except for the Background Interference Procedure-Benders, which were all evaluated by this author. Five categories of learning disability were clinically determined according to the following criteria: ( 1 ) Intellectual deficit characterized subjects who had a WISC Full Scale IQ equal to or greater than one standard deviation below the mean. ( 2 ) Emotional dysfunction was based primarily on the psychologists' clinical observations of students' behavior during testing but also on possible "signs" of emotional difficulty appearing on the Bender-Gestalt (Koppitz, 1963), as well as on the quality of WISC responses. ( 3 ) Percepmal dysfunction was based on a Bender score (Koppitz, 1963) equal to or greater than one standard
770
B. L. MALLINGER
deviation below the mean for the child's chronological age; the presence of possible "neurological indicators" (Koppitz, 1963), and/or more than two WISC Performance Scale subtest scores equal to or greater than one standard deviation below the mean, and/or Verbal IQ-Performance IQ difference of at least 12 points. ( 4 ) Any combination represented any student demonstrating more than one of the above difficulties. Where possible, every effort was made to place such a child into one of the above three groups. ( 5 ) N o problem was TABLE 1 MEANSFOR 29 ORIGINAL PREDICTORVARIABLESFOR FNE Variable Information Comprehension Arithmetic Similarities Vocabulary Digit Span Picture Completion Picture Arrangement Block Design Object Assembly Coding Verbal IQ Performance IQ Full Scale IQ VIQ-PIQ Difference
1t
6.87 9.25 8.37 6.75 7.12 5.50 8.25 5.00 8.75
2 WISC 8.82 10.54 10.82 12.36 11.09 9.64 12.09 10.45 11.18
5.25
10.45
6.62 84.62 86.00 84.00 4.62
10.82 104.73 108.72 107.45 5.64
3 9.87 11.48 9.87 14.35 10.30. 7.69 11.35 8.48 8.78 7.91 8.91 105.96 76.48 101.61 12.35
4
GROUPS* 5
Total
8.11 8.44 9.11 8.72 8.67 7.05 11.33 9.05 9.83
9.70 11.09 10.26 10.79 10.44 8.94 10.73 11.26 11.23
8.67 10.16 9.68 10.59 9.52 7.76 10.75 8.84 9.95
8.61
10.73
8.59
9.00 90.44 101.94 95.44 12.94
10.21 102.97 107.12 105.44 8.85
9.11 97.74 100.05 98.78 8.88
Bender 11.96 9.44 6.00 77.35 73.67 61.55 2.96 1.56 1.OO 0.96 1.22 0.55 2.57 1.94 0.82 5.00 4.72 3.64 5.43 3.44 1.OO 9.96 9.06 7.36 89.61 82.50 73.64 43.00 36.83 38.83 2.78 1.67 1.73 3.82 4.48 4.06 Biographical 82.25 79.55 79.52 80.17 Age (mo.) Sex 0.50 0.73 0.74 0.61 N 8 11 23 18 "Abbreviations for Bender and biographical predictors are: Total Error I = Koppitz score on Bender: Total Error I1 = Canter score on Bender: Rotations. Perseverarions. Inteerations, ~isrorcions= number of errors of each type according tb Koppitz's &em; fiiscrepancy score = discrepancy between Total Error I and mean score for subject's chronological age; Canter BIP Score I = Koppitz score on Canter's BIP; Canter Score I1 = Canter score on BIP; D-score = difference between BIP I and BIP 11; Canter Rotations and Distortions = number of errors of each type o n Background Interference Procedure; Age = chronological age in months; Sex: 0 = female, 1 = male. tGroups: 1 = intellectual deficiency; 2 = emotional; 3 = perceptual; 4 = any combination; 5 = no problem. Total Error I Total Error I1 Rorations Perseverations Integrations Distortions Discrepancy Score Canter Score I Canter Score I1 D-score Canter Rotations Canter Distortions
11.37 84.37 2.00 0.75 2.87 5.75 5.62 11.25 91.37 36.75 2.62 4.37
771
PREDICTING LEARNING DISABILITY
reserved for students showing no discernible intellectual, emotional, or perceptual problem, according to the above criteria. In essence, this group served as a control. Based on the resulting data, subjects were clinically classified into one of the above five groups. Inasmuch as the classifications were considered nominally scaled data, the usual statistical procedures for determining interrater reliability were not applicable. Thus, a simple percentage of agreement was calculated. Two judges (school psychologists who performed the testing) concurred on 95% of the ratings, indicating a very high degree of agreement across independent ratings. Twenty-nine predictor variables were derived from the psychological evaluation. These included 15 WISC, 12 Bender-Gestalt and Background Interference Procedure, and 2 biographical predictors (see Table 1 ) .
RESULTS The established battery of 29 predictor variables was reduced to a set of 10 optimal predictors on the basis of highest standardized weights (see Table 2 ) . Two significant discriminant functions were generated. The first function accounted for 53% of the variance ( x =~122.89, ~ p~ .0001), while the second 7 ~ function accounted for an additional 38% of the canonical variation ( ~ 2 = 61.63, 9 < .0002). Thus, 91% of the variance was accounted for by the two
Perceptual and Motor Skills 1977
BACKGROUND INTERFERENCE PROCEDURE AND DISCRIMINANT FUNCTION ANALYSIS I N PREDICTING CLINICALLY DETERMINED CATEGORIES OF LEARNING DISABILITY1 BARRY L. MALLINGERa Radford CoZlege Summary.-There is a need to determine the extent to which the "Background Interference Procedure" as an adjunct to the Bender-Gestalt can account for criterion variance begond that level predicted by an optimal battery. Discriminant functions empirically dassified subjects into clinical categories of learning disability. A reduced battery of intellective and visual-motor predictors generated two significant functions, accounting for 91 % of the variance. The first dimension reflected over-all intellectual functioning, the second, psychomotor skills. Empirical classification accurately categorized 71 % of all s u b jects across five criterion groups. The funcrions efficiently separated the criteria, but the six Background Interference Procedure predictor variables did not improve prediction. Implications include using the Background Interference Pr* cedure for early screening of learning disabilities and employing discriminant functions for data reduction and construct validation of teachers' and judges' ratings.
The literature indicates there have been numerous problems in diagnosing the group of children with average or near-average intelligence and with learning disabilities, language, and/or perceptual-motor deficits. While definitive diagnosis of learning disabilities is hampered by elusive criteria and by the lack of homogeneity among elements within the syndrome, many experts continue to stress the importance of careful and proper diagnostic procedures (Oettinger, 1971). Two very frequently used psychological instruments for differential diagnosis with young children are the Wechsler Intelligence Scale for Children (Wechsler, 1949) and the Bender Visual-motor Gestalt Test (Bender, 1938). While many researchers and all practitioners suggest careful scrutiny of each child suspected of learning disabilities, dinical and school psychologists can verify that such a procedure is often time-consuming and expensive. It seems obvious that the use of computers could greatly facilitate this procedure. An ideal yet feasible approach would be to combine a thorough diagnostic evaluation of the child with computerized and statistical handling of the extended data. Numerous researchers have demonstrated that clinical prediction seldom betters statistical prediction for groups (Sawyer, 1966). In recent years many re'The author wishes to acknowledge, with gratitude, Dr. Steven V. Owen, University of Connecticut, for his assistance in data collection and analysis. Appreciation is also extended to the Editors of Perceptual and Mot07 Skills, for theic helpful comments in preparing this article. 'Paper presented at the annual meeting of the Virginia Academy of Science, Madison College, Harrisonburg, Virginia, May, 1975. Address correspondence to Barry L. Mallinger, Department of Psychology, Radford College, Radford, Virginia 24142.
768
B. L. MALLINGER
searchers have called for the application of multivariate procedures to dassification and prediction problems in psychology and education (Tatsuoka, 1971). A major restriction of studies investigating classification problems is that many researchers continue to employ univariate techniques, such as simple correlations between a set of predictors and a criterion, in spite of the availability of multivariate procedures. The primary advantage of these latter techniques is that one may use a combination of predictors simultaneously and eliminate predictors which do not account for a significant percentage of unique criterion variance. Several multivariate techniques have been applied in studying the classification of learning disabled school children, but discriminant function analysis is probably the most efficient for maximalIy discriminating among diagnostic groups. Ackerman, Peters, and Dykman (1971) found discriminant functions could separate learning disabled from non-disabled children on the basis of WISC predictor variables. Eaves, Kendall, and Crichton (1972) predicted learning disabilities in kindergarten children using discriminant functions which were generated from a battery of intellective and visual-motor variables. Mallinger, Owen, McCook, and Gable (1973) demonstrated the feasibility and efficiency of using behavioral predictors in a multivariate format to separate various groups of learning disabled children. While the Bender-Gestalt is popular with psychologists (Schulberg & Tolor, 1961) and has been used in a multivariate format for differential diagnosis (Burgess, Kodanat, Ziegler, & Greenburg, 1970), the test is not consistently sensitive in screening deficits in psychomotor performance of learning disabled children. A new procedure, developed to deal with problems often encountered in using the Bender for differential diagnosis, is Canter's (1966, 1968) "Background Interference Procedure." This test has been demonstrated to be more sensitive to visual-motor problems than is the standard Bender. This technique is applicable with school-age children (Adams & Canter, 1969; Kenny, 1971) and is conducive to multivariate analysis (Adams, 1970). The major objective of this research, then, was to determine whether a set of predictor variables, derived from the Background Interference Procedure for the Bender-Gestalt, could account for criterion variance beyond that level predicted by an established battery of visual-motor and intellective variables. METHOD Of the many statistical procedures considered "multivariate," multiple discriminant analysis was used in this research (Finn, 1968). This procedure basically concerns the derivation of dimensions which separate criterion groups. This technique permits the reduction of many predictors to a single weighted composite score (or more than one weighted coefficient in the case of multiple discriminant analysis) which can discriminate among criterion groups. The
PREDICTING LEARNING DISABILITY
769
researcher derives canonical variates (dimensions or functions) which maximize differences between groups on the basis of differences within groups. On the basis of standardized weights, an established battery of predictors was reduced to an optimal set, and significant discriminant functions were generated. Variables having the largest weights were applied to the sample in order to contrast the clinical, a piori diagnoses of judges with the empirical classifications. Proportions of hits and misses in the categorization of the criterion groups were calculated (Dixon, 1968), and discriminant function centroids were then plotted to interpret the dimensions in a meaningful manner.
S~lbjecls The sample consisted of 100 elementary school children, 6 to 7 yr. old, in a suburban Northeastern community. All subjects had been nominated by their teachers to attend classes for the learning disabled, and each child was given an individual psychological evaluacion. The evaluation consisted of the Wechsler Intelligence Scale for Children, the standard Bender-Gestalt, and the Background Interference Procedure. Each child was asked to draw the Bender designs rwice, once under standard conditions and once on special paper. One-half of the children drew the Bender figures on the special paper for the Background Interference Procedure before the WISC was administered, that is, at the very start of the testing. These children received the standard Bender Following the WISC. The order was reversed for the ocher one-half of the sample. This procedure was followed to investigate whether order-of-presentation had any effect on the scores and to control for a possible practice effect. Directions given to the children for the second Bender, whether Background Interference Procedure or standard, were simply: "Remember those drawings you did before! Now, I would like for you to draw the same designs again, just like you did before." When questions regarding the special paper were raised, the examiners answered in a non-committal manner, e.g., "It's up co you," or "Just do the best you can." Each psychologist scored his own tests, except for the Background Interference Procedure-Benders, which were all evaluated by this author. Five categories of learning disability were clinically determined according to the following criteria: ( 1 ) Intellectual deficit characterized subjects who had a WISC Full Scale IQ equal to or greater than one standard deviation below the mean. ( 2 ) Emotional dysfunction was based primarily on the psychologists' clinical observations of students' behavior during testing but also on possible "signs" of emotional difficulty appearing on the Bender-Gestalt (Koppitz, 1963), as well as on the quality of WISC responses. ( 3 ) Percepmal dysfunction was based on a Bender score (Koppitz, 1963) equal to or greater than one standard
770
B. L. MALLINGER
deviation below the mean for the child's chronological age; the presence of possible "neurological indicators" (Koppitz, 1963), and/or more than two WISC Performance Scale subtest scores equal to or greater than one standard deviation below the mean, and/or Verbal IQ-Performance IQ difference of at least 12 points. ( 4 ) Any combination represented any student demonstrating more than one of the above difficulties. Where possible, every effort was made to place such a child into one of the above three groups. ( 5 ) N o problem was TABLE 1 MEANSFOR 29 ORIGINAL PREDICTORVARIABLESFOR FNE Variable Information Comprehension Arithmetic Similarities Vocabulary Digit Span Picture Completion Picture Arrangement Block Design Object Assembly Coding Verbal IQ Performance IQ Full Scale IQ VIQ-PIQ Difference
1t
6.87 9.25 8.37 6.75 7.12 5.50 8.25 5.00 8.75
2 WISC 8.82 10.54 10.82 12.36 11.09 9.64 12.09 10.45 11.18
5.25
10.45
6.62 84.62 86.00 84.00 4.62
10.82 104.73 108.72 107.45 5.64
3 9.87 11.48 9.87 14.35 10.30. 7.69 11.35 8.48 8.78 7.91 8.91 105.96 76.48 101.61 12.35
4
GROUPS* 5
Total
8.11 8.44 9.11 8.72 8.67 7.05 11.33 9.05 9.83
9.70 11.09 10.26 10.79 10.44 8.94 10.73 11.26 11.23
8.67 10.16 9.68 10.59 9.52 7.76 10.75 8.84 9.95
8.61
10.73
8.59
9.00 90.44 101.94 95.44 12.94
10.21 102.97 107.12 105.44 8.85
9.11 97.74 100.05 98.78 8.88
Bender 11.96 9.44 6.00 77.35 73.67 61.55 2.96 1.56 1.OO 0.96 1.22 0.55 2.57 1.94 0.82 5.00 4.72 3.64 5.43 3.44 1.OO 9.96 9.06 7.36 89.61 82.50 73.64 43.00 36.83 38.83 2.78 1.67 1.73 3.82 4.48 4.06 Biographical 82.25 79.55 79.52 80.17 Age (mo.) Sex 0.50 0.73 0.74 0.61 N 8 11 23 18 "Abbreviations for Bender and biographical predictors are: Total Error I = Koppitz score on Bender: Total Error I1 = Canter score on Bender: Rotations. Perseverarions. Inteerations, ~isrorcions= number of errors of each type according tb Koppitz's &em; fiiscrepancy score = discrepancy between Total Error I and mean score for subject's chronological age; Canter BIP Score I = Koppitz score on Canter's BIP; Canter Score I1 = Canter score on BIP; D-score = difference between BIP I and BIP 11; Canter Rotations and Distortions = number of errors of each type o n Background Interference Procedure; Age = chronological age in months; Sex: 0 = female, 1 = male. tGroups: 1 = intellectual deficiency; 2 = emotional; 3 = perceptual; 4 = any combination; 5 = no problem. Total Error I Total Error I1 Rorations Perseverations Integrations Distortions Discrepancy Score Canter Score I Canter Score I1 D-score Canter Rotations Canter Distortions
11.37 84.37 2.00 0.75 2.87 5.75 5.62 11.25 91.37 36.75 2.62 4.37
771
PREDICTING LEARNING DISABILITY
reserved for students showing no discernible intellectual, emotional, or perceptual problem, according to the above criteria. In essence, this group served as a control. Based on the resulting data, subjects were clinically classified into one of the above five groups. Inasmuch as the classifications were considered nominally scaled data, the usual statistical procedures for determining interrater reliability were not applicable. Thus, a simple percentage of agreement was calculated. Two judges (school psychologists who performed the testing) concurred on 95% of the ratings, indicating a very high degree of agreement across independent ratings. Twenty-nine predictor variables were derived from the psychological evaluation. These included 15 WISC, 12 Bender-Gestalt and Background Interference Procedure, and 2 biographical predictors (see Table 1 ) .
RESULTS The established battery of 29 predictor variables was reduced to a set of 10 optimal predictors on the basis of highest standardized weights (see Table 2 ) . Two significant discriminant functions were generated. The first function accounted for 53% of the variance ( x =~122.89, ~ p~ .0001), while the second 7 ~ function accounted for an additional 38% of the canonical variation ( ~ 2 = 61.63, 9 < .0002). Thus, 91% of the variance was accounted for by the two
