A RECONCEPTUALIZATION AND REVISED SCORING PROCEDURE FOR THE ITPA BASED ON MULTIVARIATE ANALYSIS OF THE ORIGINAL NORMATIVE DATA GARYCRONKHITE and KANDACEPENNER
California State University, San lose, California This research was conducted to (1) determine whether a factor analysis of the original normative ITPA data would support the dimensions of Osgood's psycholinguistic model which provides its theoretical base, (2) investigate the need for constructing psycholinguistie profiles based upon this empirical factor structure rather than upon a priori theoretical "facet structure", and (3) devise a weighted multiple regression formula for estimating psycholinguistic age based on subtest correlations with chronological age for the original normative group. Whereas five factors were anticipated, one of these split into two components. The factors represented the (1) perceptual process at the representation level, accounting for 8~ of the variance; (2) perceptual process at the integration level, 6.5~; (3) organizing rocess at the representation level, 25~; and (4) expressive process at the integration ~evel, 11~9 The exp ected integration-organizingfactor split into (5) auditory memory, 9~;, and (6) visual memory, 9~. As anticipated, no emergent factor represented the expressive process at the representation level. The emergent factor structure strongly su~ggests that the use of factor profiles would be more meaningful than the use of subtest profiles. A multiple regression formula is presented for computing psycholinguistic age equivalents. Kirk (1974) has recently criticized previous factor analytic studies of the Illinois Test of Psycholingnistic Abilities (ITPA) on three grounds: (1) subjects have been poorly selected and tests poorly administered, (2) the studies have not proceeded from "a careful a priori theoretical framework," and (3) the studies have not included external criterion tests to establish construct validity of the ITPA subtests. The present study was designed to resolve the first two problems by (1) performing factor analyses on data gathered under Kirk's supervision, the data on which the test norms are based (see Paraskevopoulos and Kirk, 1969) and (2) proceeding from Osgood's (1957a, b, 1963) theoretical model, the model used by Kirk, McCarthy, and Kirk (1968) in developing both the experimental and revised editions of the test. Kirk's third criticism, however, calls for some discussion. Many writers have made the point that construct validation of the ITPA by factor analysis requires inclusion of external criterion tests (see Carroll, 1972; Proger, Cross, and Burger, 1973; Hare, Hammill, and Bartel, 1973; Newcomer et al., 1975). Numerous studies have been conducted without employing this technique. 506
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
Surveys of that research are available in Meyers (1969), Ryckman and Wiegerink (1969), and Sedlack and Weener (1973). To our knowledge only three studies have been conducted which have included external criterion variables in the factor analyses (see Uhl and Nurss, 1970; Hare et al., 1973; Newcomer et al., 1975). Generally speaking, this criticism is well taken. Of course it is important to establish construct validity, to identify the unique ability tapped by each subtest. However, establishing construct validity is not the only purpose of factor analysis. It is also important to determine the internal structure of a test such as the ITPA, so as to identify interrelationships among the subtests. Such interrelationships are important to the clinician attempting to interpret subtest scores and design remedial programs. Unfortunately, both purposes cannot be served by the same factor analysis. A factor analysis is performed on a specific data domain. Introducing external criterion tests into a data domain will alter the factor structure of that domain. For example, suppose a data domain has been largely accounted for by tffo factors of roughly equal strength. One could easily inflate the percentage of variance accounted for by one factor by adding variables likely to load on that factor. Alternatively, one could split one of the factors by adding new variables. That fact is evident in the studies which have used external criterion tests. Uhl and Nurss (1970), for example, included 53 variables in their analyses, 10 of which were drawn from the ITPA. Their first three factors contained no loadings of 0.40 or better on any ITPA subtest. Similarly, Hare et al. (1973), who used 10 external criterion tests in addition to six of the ITPA subtests, found a first factor which accounted for 20~ of the total variance but did not load to criterion on any ITPA subtest. Beyond that first factor, each of the six ITPA subtests did load on a separate factor. The important point is that, while factor analyses which include external criterion tests in addition to ITPA subtests are necessary to determine construct validity of the subtests, such studies are not factor analyses of the ITPA. It was the purpose of this study to provide a factor analysis of the internal structure of the ITPA. No external criterion tests were included. R A T I O N A L E FOR THE STUDY The present research developed as an attempt to answer four questions regarding the interpretation and validity of the Revised Illinois Test of Psycholinguistic Abilities as it is presently administered and scored. Two of those questions involve the theoretical basis for the test, and two involve its scoring.
Questions, Regarding Theoretic Rationale The authors of the ITPA adopted Osgood's (1957a, b) model of psycholinguistic functions, which he elaborated more recently in his 1962 keynote address as president of the American Psychological Association (see Osgood, CF,ONKI-IITE, PENNER" ITPA
507
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
1963). The model predates transformational grammar, and has not been updated to take such matters into account. It has been criticized on that and other grounds. See, for example, the debate developing from Osgood (1963) through Fodor (1965) back to Osgood (1966). However, this is the model upon which the ITPA is based. In the 1963 work, Osgood describes three functional levels, but we need not be concerned with the lowest level since it involves only reflex-type direct stimulus-response connections which do not require integration and coordination. The upper two levels are those of integration and representation, which seem to parallel the ITPA automatic and representation levels. Integration is the level at which incoming stimuli are organized into a perceptual pattern and at which individual muscle responses are coordinated into patterns of activity. Representation is distinguished in that it is the only level at which internal mediating responses intervene between input and output. The internal mediating response is at the heart of Osgood's theory of meaning. It is a covert response to incoming stimuli, but it has the characteristics of a stimulus in itself since it in turn occasions overt behavior. Kirk et al. (1968, p. 7) also seem to recognize that "meaning" or symbolic activity constitutes the defining characteristic of the representation level and, thus, representation is the level at which language qua language must operate. They also describe the three processes at each level consistent with Osgood: In analyzing behavior which occurs in the acquisition and use of language, three main processes are considered: (a) the receptive process, that is, the ability necessary to recognize and/or understand what is seen or heard; (b) the expressive process, that is, those skills necessary to express ideas or to respond either vocally or by gesture or movements; (c) an organizing process which involves the internal manipulation of percepts, concepts, and linguistic symbols. It is a central mediating process elicited by the receptive process and preceding the expressive process. (p. 7) The ITPA authors add to these two dimensions of level and process a third, which they term channel. They propose to test two channels-the auditoryverbal arid the visual-motor. One problem is that this suggests that auditory input yields only verbal response, and visual input yields only motor response. In fact, that is exactly how they diagram these channels in their picture of the model (p. 8). Now no one would seriously pursue such a proposition; obviously, one frequently responds manually to auditory input and verbally to visual input. Clearly, there are a variety of input channels and a variety of output channels with the organizing process matching given input regardless of channel to appropriate output regardless of channel. A similar assumption seems to pervade the test itself. Specifically, although the subtest titles and the authors themselves (pp. 9-13) suggest that the visual and auditory tests generally measure parallel functions, in fact the auditory tests involve language almost without exception, whereas the authors seem to have gone to great pains to exclude language from the 508 1ournal of Speech and Hearing Research
18 506-520 1975
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
visual tests. Carroll (1972) makes the point that the ITPA may be misnamed since it tests so many nonlanguage abilities. This last statement requires a little qualification. All the tests involve language in the directions, and the visual tests sometimes require a minimal language output, but the functions measured by the auditory tests are almost exclusively linguistic and those measured by the visual tests are almost exclusively nonlinguistic. Thus, the channel dimension in the test is almost completely confounded with the level dimension (remembering that the distinction between the two levels depends on the involvement of language). The few instances in which this is not true seem inadvertent and only serve to confuse the picture further. We anticipated that five factors would emerge from the analysis, and that these factors would correspond to the perceptual and organizing processes at the representation level and to the perceptual, organizing, and expressive processes at the integration level. We anticipated no factor corresponding to the expressive process at the representation level because we found no subtest which appeared to measure expressive skills at that level. (We have substituted perceptual process for reception process since reception is more frequently used in the psychological literature to refer to simple transmission of sensation to the cortex without organized patterning.) We anticipated, then, that the 12 ITPA subtests would distribute their high loadings among those five factors as indicated in Table 1. Following the TABLE 1. Anticipated factors for third- and fourth-grade groups.
Level Representation
Perception auditory closure sound blending
Process Organizing auditory reception grammaticclosure auditory association
Expression none
Integration
visual closure visual association
visual memory auditory memory
manual expression verbal expression
finding of Ryckman and Wiegerink (1969) that ITPA factors increase in number and stabilize at the upper age levels, we anticipated this structure only for the upper age group. Since these expectations are based on analysis of the items within each subtest, the subtest titles in many cases suggest they may be misplaced. That requires some explanation. Integration-Perception. Skills necessary to serve this function involve the reception and patterning of nonverbal stimuli, or nonverbal pattern recognition (whether visual or auditory). The subtests assigned to this factor are all visual because no ITPA subtest measures auditory nonverbal pattern recognition. Looking at a picture of a dog and then finding all the dogs in a Complex visual field (as in visual closure) directly tests nonverbal pattern recognition, as does associating a pillow with a bed rather than with a cup, refrigerator, or desk (as in visual association), although this tests recognition of functional .CB.ONKI-IITE, PENNER: ITPA
509
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
patterns. Matching a picture of a chair with a picture of another type of chair (visual reception) rather than with a cushion, table, or person is another type of pattern recognition. However, while success on such an item does not require use of the word chair as an internal mediating response, use of the word certainly facilitates success. In addition, such a test should load on a short-term memory factor and is very similar to the visual association subtest. Consequently, we anticipated that the visual reception subtest would split its loadings among several factors. Representation-Perception. Skills necessary to serve this function involve verbal pattern recognition. Completion of the verbal patterns "Da_ _y" (auditory closure) and "s_ _tar" (sound blending) tests this function quite directly. For the reason mentioned earlier, we also anticipated that visual reception would load moderately on this factor. Integration-Organizing. Skills necessary to serve this function involve the retention of a perceived pattern to serve as a bridge between pattern recognition and coordinated response. The visual and auditory sequential memory tests measure this function quite directly. We anticipated that the visual test would have the higher loading, since it is less likely to be contaminated by language than is the auditory test. If auditory sequential memory were tested by the use of nonsense syllables rather than digits, as suggested by Emerick and Hatten (1974), it would also be less contaminated. Representation-Organizing. Skills necessary to serve this function involve the use of language as an organizing device. Possession of such skills is indicated by an extensive receptive vocabulary and an understanding of the complex rules of phonemics, morphemics, and syntax. A "yes" answer to the question "Do migratory birds traverse?" (auditory reception) suggests considerable development of such skills. Correct transformation of "Here is a bed" into "Here are two beds" (grammatic closure) indicates an understanding of the rules, although the test does not necessitate a very extensive vocabulary. "Houses have architects; books have authors" (auditory association) requires considerable use of language as an organizing device. Integration-Expression. Skills necessary to serve this function involve production of a relatively extensive and coordinated series of responses which do not require language. The test of manual expression measures these skills quite directly. But why did we anticipate a high loading for verbal expression on a nonverbal factor? The answer is that, while the test of verbal expression requires minimal language, it does not measure language skills except at the lowest levels. The secret is in the scoring, which does not reflect lexical richness nor phonemic, morphemic, and syntactic complexity of the expression produced. Consider the example of a child asked to describe an envelope. A child who says "You put paper in it to mail" gets exactly the same credit as the one who says "The primary function of this obiect is to enclose stationery for transportation by the U. S. Postal Service." Points are credited for the number of characteristics mentioned regardless of the extensiveness of the vocabulary or the syntactic complexity of the utterance. What is being scored 510 Journal of Speech and Hearing Research
18 506-520 1975
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
in the test of verbal expression is essentially what is being scored in the test of manual expression. So long as the child has minimal vocabulary and syntax, language is irrelevant to the scoring. (Carroll, 1972, makes this point well. ) Representation-Expression. Skills necessary to serve this function involve expressive vocabulary and verbal output which demonstrates phonemic, morphemic, and syntactic complexity. Given the nature of the scoring of verbal expression, there is no expressive test which adequately measures this function. If there were such a test, verbal expression might correlate with it to some extent, but a minor loading of one test is not likely to produce a separate factor. Obviously our assignment of the subtests to the six cells of Osgood's model differs drastically from the conceptualization described by the test's authors. Most of the differences stem from our assumption that any subtest which measures verbal skills to any appreciable extent must be assigned to the representational level and a predominantly nonverbal test cannot be assigned to that level. Another difference is our assumption that tests which measure pattern recognition should be assigned to what we have termed the perceptual process, reserving the term organizing process for those functions which mediate between perception and expression. The remaining difference, the assignment of the auditory reception subtest to the representation-organizing cell, reflects our judgment that this subtest measures verbal mediation rather than verbal pattern recognition.
Questions Regarding Scoring Procedures On the basis of this analysis we further concluded that the customary computation of the psycholinguistic profile is probably misleading, for two reasons. First, when two or more subtests measure the same factor, use of individual subtest scores may disguise a more general problem. Low scores on a number of related subtests may be treated as independent, single-skill deficiencies when in fact they may represent a general deficiency in one of the major functions we have described. Second, to the extent that a single subtest measures more than one function, the score on that subtest may be conceptually meaningless. Consequently, we believe that t h e emergence of factors which do not parallel the subtests is evidence of the need to compute factor scores, which will constitute a more meaningful psycholinguistic profile. We also believe the composite Psycholinguistic Age (PLA) equivalents are not as accurate as they can be. The ITPA composite score consists of a straight linear addition of the subtest raw scores. This composite is then referred to age norms to determine the composite PLA. The problem is that this procedure treats all subtest scores as if they are equally correlated with age, which seems unlikely. The better procedure would be to determine the correlation of each subtest with chronological age for the normative group, and then perform a multiple regression analysis, which yields a weighted CRONKttITE, PENNER: ITPA
511
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
formula for the prediction of chronological age. A further option would then be to divide this predicted age by actual chronological age and multiply by 100 to obtain a Psycholinguistic Quotient similar to an IQ score. METHOD The data analyzed were those from the original normative group for the Revised ITPA. Consequently, extensive description of the subjects and test administration procedures are available in Paraskevopoulos and Kirk (1969).
Subjects The scores were elicited from 961 children ranging in age from two years seven months to 10 years one month. They are described as having "average intellectual functioning . . . average school achievement . . . sensorimotor integrity . . . at least average characteristics of personal-social adjustment" and being "from English-speaking families" (p. 53). They were drawn from four Illinois communities near Champaign and from Madison, Wisconsin. The proportion of males to females at each age level was approximately 50%, but the percentage of black children was only 4~;, well below the national average. Socioeconomic status indicators approximated national norms. We divided the total sample into four subsamples on the basis of age for the purpose of performing four-factor analyses. The preschool group consisted of 220 children ranging in age from two years seven months to four years one month. The advanced preschool and kindergarten group included 246 children with an age range from four years seven months to six years one month. The first- and second-grade group included 247 children from six years seven months to eight years one month, while 248 children in the third and fourth grades ranged from eight years seven months to 10 years one month. We used all subjects in the multiple regression analysis.
Test Administration Testing was conducted from April through July of 1967. All school-age children were tested in their schools, in special test rooms, while preschool children were generally tested at home. Most examiners were graduate students in psychology and education who had previous training in administering the Stanford-Binet, and most of them had also been trained in administering the experimental version of the ITPA. All were given specific training in administering the Stanford-Binet and the Revised ITPA and were frequently observed and advised.
Statistical Analyses The variables analyzed in the present study were the Revised ITPA subtest 512 1ournal of Speech and Hearing Research
18 506-520 1975
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
scores and chronological age in months. Six analyses were performed. Four of these were varimax factor analyses (orthogonal rotations) of the data for children in the four restricted age ranges already described. All analyses were performed by means of the STATPAK programs run at the Computer Center at California State University, San Jose. The variables involved in these analyses were the 12 ITPA subtest scores. The eigenvalue was initially set at 1.0 as has become more or less customary (see Kaiser, 1960; Harman, 1967). When we observed that our predicted short-term memory factor was instead represented by two factors (auditory and visual memory) we dropped the eigenvalue to 0.75, allowing six factors to emerge. A stepwise multiple regression analysis was performed using chronological age as the dependent variable and the 12 subtest scores as the predictor variables. RESULTS The first factor analysis at each level was performed with the eigenvalue set at the customary 1.0. Thus, with 12 variables, no factor was extracted which accounted for less than 8.3~ of the total variance. Only one factor emerged from the analysis of scores of the preschool group. That factor accounted for 49~ of the total variance. Subtest loadings on that factor were as follows: auditory reception, 0.78; visual reception, 0.64; auditory association, 0.80; visual association, 0.75; verbal expression, 0.68; manual expression, 0.66; grammatic closure, 0.85; visual closure, 0.69; auditory memory, 0.59; visual memory, 0.59; auditory closure, 0.71; and sound blending, 0.59. Since only one factor was extracted, these are unrotated loadings. Three factors emerged from the analysis of advanced preschool and kindergarten scores. Factor I accounted for 39~ of the total variance, Factor II for 10~, and Factor III for 9~. The subtest loadings of 0.30 or better on those factors are displayed in Table 2. TABLE 2. Varimax factor loadings for preschool and kindergarten (eigenvalue = 1.0).
Factors
I
Factor Number H
IH
Auditory reception Visual reception Auditory association Visual association Verbal expression Manual expression Grammatie closure Visual closure Auditory memory Visual memory Auditory closure Sound blending
0.59 0.73 0.64 0.73 0.64 0.67 0.51 -
0.33 0.76 0.80 0.50 -
0.45 0.79 0.62
CRONKHITE, PENNER: ITPA
513
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
Three factors also emerged from the analysis of scores of first- and secondgrade children. Variances accounted for were as follows: Factor I, 29~; Factor II, 10'/;; and Factor III, 10~. Subtest loadings for this analysis are displayed in Table 3. T ~ L ~ . 3. V a r i m a x f a c t o r l o a d i n g s f o r first a n d s e c o n d g r a d e s ( e i g e n v a l u e
Factors
I
Auditory reception Visual reception Auditory association Visual association Verbal expression Manual expression Grammatie closure Visual closure Auditory memory Visual memory Auditory closure Sound blending
=
Factor Number H
0.35 0.51 0.32 0.71 0.76 0.44 0.48 -
-0.49 - 0.53 - 0.41 -0.64 -0.73 -0.71
1.0).
Ill --0.46 - 0.36 -0.38 -0.54 -0.75 -
Four factors were extracted in the analysis of scores of third- and fourthgrade children. Variances accounted for were as follows: Factor I, 25~; Factor II, 11~; Factor III, 10~; and Factor IV, 9~. Subtest loadings are displayed in Table 4. TABLE 4. V a r i m a x f a c t o r l o a d i n g s f o r t h i r d a n d f o u r t h g r a d e s ( e i g e n v a l u e
Factors Auditory reception Visual reception Auditory association Visual association Verbal expression Manual expression Grammatie closure Visual closure Auditory memory Visual memory Auditory closure Sound blending
I 0.33 0.50 0.58 0.72 0.73 0.36 0.48 -
Factor Number H III 0.53 0.40 0.46 0.50 0.66 0.73
=
1.0).
IV
-0.41 0.89 -
0.37 0.39 -0.31 0.84 -
We anticipated the emergence of five specific factors consisting of a specific pattern of subtest loadings for the analysis of scores of third- and fourth-grade children only. When the analysis using an eigenvalue of 1.0 produced only four factors, and when we observed that the expected short-term memory factor had split into auditory and visual memory, we adjusted the eigenvalue to 0.75 to see if the next two factors would be those we anticipated. The dearest display of the results seems to be in Table 5, which is patterned directly after 514
1ournal o[ Speech and Hearing Research
18
506-520
1975
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
l:m
i
9
~
~
.
0
CI'~ONg_KrrE, PENNER: ITPA 515
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
Table 1 in which our expectations were displayed. Only subtest loadings greater than 0.50 are included. This pattern of results is remarkably similar to those anticipated. The exceptions are these: (1) the predicted integration-organizing factor (short-term memory) was represented instead by separate auditory and visual memory factors; (2) the visual association subtest, predicted to load above 0.50 on the integration-perception factor, instead showed moderate loadings on the representation-organizing, visual memory, and integration-expression factors; and (3) auditory association, while showing the high loading predicted on the representation-organizing factor, was contaminated by a moderate loading on the integration-expression factor. A multiple regression formula using subtest scores for grammatic closure, auditory association, and visual closure alone produced a multiple correlation coefficient of +0.954 with chronological age, accounting for 91.0% of the variance. The intercept for that formula was 26.48, and the regression coefficients were 1.27, 0.70, and 0.73, respectively. Adding the remaining nine subtests to that formula raised the multiple/t only to 0.963 and increased the variance accounted for by only 1.7~. In fact, the grammatic closure subtest alone produced a correlation coefficient of 0.934, accounting for 87.2~ of the variance, using an intercept of 31.22 and regression coefficient of 2.67. DISCUSSION We can now consider the implications of each of these findings in turn. Some of the implications are for future theory and research; others are relevant to practical procedures of clinicians using the test for diagnosis. Use of Composite PLA
The findings suggest that the use of composite scores is likely to be misleading. Just as adding a child's IQ and height will produce a meaningless number, so will the addition of scores on subtests which measure different and uncorrelated factors or psycholinguistic abilities. The authors of the test argue that such addition is justified by the fact that the subtests are correlated with one another. We contend that those intercorrelations are too weak to provide such justification. No two subtests have more than 16~ of their variance in common. Some subtest intercorrelations are simply negligible. No subtest intercorrelations approach 0.50, which we demanded as the minimum correlation of a subtest with a factor. (Subtest intercorrelations at various age levels are reported by Paraskevopoulos and Kirk, 1969, Appendix). Actually, the composite score is almost always used only as an intermediate step to computing composite Psycholinguistic Age. Our stepwise multiple regression analysis provided two rather simple and accurate formulas for relating subtest scores to chronological age. The simpler of these, using only the grammatic closure raw score, yields estimated PLA scores which showed 516 1ournal 6f Speech and Hearing Research
18 506-520 1975
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
a 0.934 coefficient of correlation with chronological age. That formula is simply estimated PLA = 31.22 + 2.67 (GC) where GC represents the grammatic closure subtest raw score. The result is estimated PLA expressed in months. The more accurate estimate, which yielded a slightly better 0.954 coefficient of correlation with chronological age, is produced by the following formula: estimated PLA = 26.48 + 1.27 (GC) +0.70 (AA) +0.73 (VC) where AA and VC represent the auditory association and visual closure subtest scores, respectively. Either of these formulas will provide the clinician with quite accurate PLA estimates, and both avoid the demonstrably false assumption of subtest additivity.
Use of Psycholinguistic Profile As mentioned earlier, the use of subtest profiles can cause at least two types of problems. First, when two or three subtests show high loadings on the same factor, low scores on those subtests suggest a general deficit in the function represented by that factor. The clinician who looks at the subtests individually may overlook the general functional deficiency and try instead to remedy the specific skills represented by the subtests. Practice on those specific tasks may allow the child to show a spurious and artificial improvement in his subtest scores without remedying tile general deficiency. Alternatively, if the general deficiency is too serious, and if the specific skill deficiencies are only symptomatic rather than causative, practice on the specific skills may produce little improvement in subtest scores so long as the general deficiency is not remedied. For example, a pattern of low scores on the auditory reception, grammatic closure, and auditory association subtests suggest that the child may have difficulty comprehending or applying the rules of the language. Practice on the specific language tasks involved in those subtests may or may not improve his scores. Either way, a clinician who recognizes the general problem should be able to give the child more real help by devising imaginative means of calling his attention to the language rules which have eluded him and teaching him to apply those rules. Second, there are two subtests, visual reception and visual association, which appear to be very ambiguous in that they show moderate loadings on three factors and no high loadings on any single factor. Low scores on these tests can only be interpreted to the extent that they participate in a pattern of low scores. Thus, low scores on these two subtests coupled with a pattern of low scores on the representation-organizing subtests would be interpretable, but they would be ambiguous taken alone. CI:iONKHITE, PENNER: ITPA
517
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
How, then, can the clinician use the factor-analytic results to construct a more meaningful profile? Our suggestion at this point is to reorganize the profile chart so that those subtest scores which reflect a given factor are adjacent to one another and separated from other factors by some visual mnemonic device such as colored or dotted lines. This would make it much easier to observe a general factor deficit. The two ambiguous subtests (visual reception and association) could be displayed last and interpreted only in combination with other patterns. Eventually, it may be possible to calculate factor PLA scores, but that must await the results of further research.
Implications of the Model As noted earlier, we have interpreted many of the subtests rather differently from the ways the authors of the ITPA interpret them. The findings appear quite consistent with our interpretations. If the reader is willing to accept our reinterpretation and reassignment of subtests to the cells of Osgood's model, several implications follow. The most glaring of these is the absence of any subtest which adequately measures expression at the representation level. (Carroll, 1972, also noted this problem. ) It seems strange, indeed, that the ITPA, which is used so widely to diagnose expressive language disorders, contains no subtest which measures that function directly. The research implication is dear: we need a subtest which directly tests a child's facility in applying the phonemic, morphemic, syntactic, and semantic rules in oral expression. The dinicaI implication is that the ITPA in its present form must be supplemented by some other test of that function. A second problem is that some of the tests which are generally taken to measure the reception or organization processes do not appear to measure the processes to which they are customarily assigned. The clinician should be especially suspicious of the auditory closure, visual closure, and auditory reception, since they appear to be quite misleading. Auditory closure is assumed to be a test of an organizing process at the automatic (nonlangnage) level when in fact it depends on linguistic (semantic) knowledge of the child who perceives and effects closure on the word presented. Thus it seems more appropriately considered a perceptual test at the representational level. The sound blending subtest certainly asks the child to do what its name implies, but seems to be better placed in the model as a test of verbal pattern recognition (that is, verbal perception). While auditory closure and sound blending depend on perception of linguistic units, auditory reception requires an ability not only to perceive but also to organize linguistic stimuli, thus making it a test of the organizing rather than just the perceptual process. Finally, the visual, auditory, motor, and verbal channel labels, as applied to the subtests, are misleading in that they imply parallelism where none exists. This is confusing because the differences between what are supposedly subtests of different channels are in fact differences between subtests at different 518 Journal of Speech and Hearing Research
18 506-520 1975
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
levels distinguished by language involvement or the lack of it. For example, visual reception and auditory reception are supposed to test the perceptual process at the representational level through two separate channels. We found that visual reception, for example, loads to some extent on a factor represented most clearly by some auditory subtests, apparently because of its language involvement. Verbal expression shares a very high integration-expression loading with manual expression, indicating that they measure the same thing, perhaps the willingness to express oneself. Clinically, the implications are clear. A discrepancy between auditory and visual pairs of subtests (usually high visual and low auditory) does not automatically indicate a high visual and low auditory ability in a specific function. This is because subtests with parallel labels do not necessarily measure parallel functions. Generally, the clinician would be well advised to ignore those labels and look instead at the factor labels which seem far more descriptive. For the clinician who wishes to consider the results of an ITPA in light of the author's reinterpretation, a reorganization of the psycholinguistic profile may be helpful. The results of the present study indicate that the abilities measured by the ITPA subtests fall into five ability categories or factors. Therefore, when listing or plotting the subtest scores, it would be beneficial to use a method which would constitute a visual reminder of the relatedness of certain of the subtests. One such method is to use the model in Table 5 recording the subtest score beside its name. Another procedure utilizes the Profile of Abilities graph. The scaled scores are plotted on the graph as usual but related scores are plotted in the same color. Whatever method is used, it will be apparent if the child is low on one or more subtests in each factor. If the child scores low on, for example, auditory reception, grammatic closure, and auditory association, all of which load on the same factor, it would appear that the child has an overall disability in the organizing process at the representational level, in other words, in organization of linguistic stimuli. If, however, he scores low on only one of the above subtests, he may have a less general disability. Such information regarding the relatedness of ITPA subtests would appear to be important in the planning of remedial procedures. CONCLUSION In short, these multivariate analyses of the ITPA subtests provide considerable support for Osgood's model, but little support for the ITPA authors' conception of their test or its relation to that model. Further, the analyses strongly suggest that the Psycholinguistic Profiles and Psycholinguistic Age scores would be considerably more meaningful and useful if the scoring procedures were revised to reflect the relationships among the subtests which are clearly observable in the normative data. The ITPA emerges as a useful test, but the conceptualization of the test and the scoring procedures recommended in the examiner's manual emerge as misleading in a number of respects.
CRONKHITE, PENNER: I T P A
519
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx
ACKNOWLEDGMENT Gary Cronkhite is currently affiliated with the Department of Rhetoric, University of California at Davis. Kandace Penner is in the Department of Education, University of California at Berkeley. REFERENCES
CARROLL, J. B., Critique of the Illinois Test of Psycholinguistic Abilities. In O. K. Buros (Ed.), Seventh Mental Measurement Yearbook. Highland Park, N.J.: Gryphon, 819-823 (1972). EMEmCK, L. L., and HXTTEN, J. T., Diagnosis and Evaluation in Speech Pathology. Englewood Cliffs, N.J.: Prentice-Hall (1974). FODOR, J. A., Could meaning be an rm? J. verb. Learn. verb. Behav., 4, 73-81 (1965). H,~mE, B. A., H,~M~ILL, D. D., and BARREL, N. R., Construct validity of selected subtests of the ITPA. Except. Child., 40, 13-20 (1973). H.~,MAN, H. H., Modern Factor Analysis. Chicago: Univ. of Chicago Press (1967). K~SER, H. F., The application of electronic computers to factor analysis. Educ. psychol. Measur., 20, 141-151 (1960). Kn~, S. S., The ITPA and research. Paper presented at the Annual Convention of the American Speech and Hearing Association, Las Vegas, Nevada, (November 1974). KIRK, S. S., MCCARTaY, J. j., and Kiax, W. D., Examiner's Manual, Illinoi~ Test of Psycholinguistic Abilities. (rev. ed.) Urbana: Univ. of Illinois (1968). MEYERS, C. E., What the ITPA measures: A synthesis of factor studies of the 1961 edition. Educ. psychol. Measur., 29, 867-876 (1969). NEWCOMER, P., HArm, B. A., HAM~ILL, D. D., and MCGETTIGAN, J., Construct validity of the ITPA. 1. learn. Disabil., 8, 32-43 ( 1975 ). OsGooo, C. E., A behavioristic analysis of perception and language as cognitive phenomena. In Contemporary Approaches to Cognition. Cambridge, Mass.: Harvard Univ. Press, 75118 (1957a). OSGOOD, C. E., Motivational dynamics of language behavior. In M. R. Jones (Ed.), Nebraska Symposium on Motivation. Lincoln: Univ. of Nebraska Press, 348-424 (1957b). OSGOOD, C. E., On understanding and creating sentences. Am. Psychol., 18, 735-751 (1963). OsGooo, C. E., Meaning cannot be an rm?. 1. verb. Learn. verb. Behav., 5, 402-407 (1966). PAaASr~VOPOULOS,J. N., and Kxmr S. A., The Development and Psychometric Characteristics of the Revised Illinois Test of Psycholinguistic Abilities. Urbana: Univ. of Illinois Press (1969). PROGER,B. B., CROSS, L. H., and BURGER,R. M., Construct validation of standardized tests in special education: A framework of reference and application to ITPA research ( 1967-1971 ). In L. Mann and D. Sabatine ( Eds. ), The First Review of Special Education. Philadelphia: Buttonwood Farms (1973). RYCKMAN, D. B., and WIEGERINK,R., The factor of the Illinois Test of Psyeholinguistic Abilities: A comparison of eighteen factor analyses. Except. Child., 36, 107-113 (1967). SEDLAK, R. A., and WEENEn, P., Review of research on the Illinois Test of Psycholinguistic Abilities. In L. Mann and D. Sabatine (Eds.), The First Review of Special Education. Philadelphia: Buttonwood Farms, 113-163 (1973). UHL, N. P., and Ntrass, j. R., Socioeconomic level styles in solving reading related tasks. Reading Res. Q., 3, 452-485 (1970). Received July 1, 1974. Accepted April 20, 1975.
520 lournal of Speech and Hearing Research
18 506-520
1975
Downloaded From: http://jslhr.pubs.asha.org/ by a University of Minnesota, Minneapolis - Library User on 02/01/2018 Terms of Use: http://pubs.asha.org/ss/rights_and_permissions.aspx