A confirmatory factor analysis of the WMS-III in a clinical sample with crossvalidation in the standardization sample

Transcription

1 Archives of Clinical Neuropsychology 18 (2003) A confirmatory factor analysis of the WMS-III in a clinical sample with crossvalidation in the standardization sample D. Bradley Burton a,, Joseph J. Ryan b, Bradley N. Axelrod c, Tony Schellenberger a, Heather M. Richards a a Division of Social Sciences, Indiana University Southeast, 4201 Grantline Road, New Albany, IN , USA b Central Missouri State University, Warrensburg, MO, USA c John D. Dingell Veterans Affairs Medical Center, Detroit, MI, USA Accepted 25 February 2002 Abstract A maximum likelihood confirmatory factor analysis (CFA) of the Wechsler Memory Scale-III (WMS-III) was performed by applying LISREL 8 to a general clinical sample (n = 281). Analyses were designed to determine which of seven hypothesized oblique factor solutions could best explain memory as measured by the WMS-III. Competing latent variable models were identified in previous studies. Results in the clinical sample were crossvalidated by testing all models in the WMS-III standardization samples (combined n = 1,250). Findings in both the clinical and standardization samples supported a four-factor model containing auditory memory, visual memory, working memory, and learning factors. Our analysis differed from that presented in the WMS-III manual and by other authors. We tested our models in a clinical sample and included selected word list subtests in order to test the viability of a learning dimension. Consistent with prior research, we were also unable to empirically support the viability of the immediate and delayed memory indices, despite allowing the error terms between the immediate and delayed memory subtests to correlate Published by Elsevier Ltd on behalf of National Academy of Neuropsychology. Keywords: WMS-III; Confirmatory factor analysis; WMS-R Corresponding author. Tel.: ; fax: address: dburton@ivs.edu (D. Bradley Burton) /$ see front matter 2002 National Academy of Neuropsychology. PII: S (02)00149-X

2 630 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) Introduction The Wechsler Memory Scale-III (WMS-III) is the most recent version of the Wechsler Memory Scale (The Psychological Corporation, 1997). It contains 11 primary subtests and 8 primary indexes that include auditory immediate, visual immediate, immediate memory, auditory delayed, visual delayed, auditory recognition, general memory, and working memory. These eight indexes are constructed to provide summary measures that are thought to reflect specific aspects of memory as measured by the WMS-III. There are also a number of supplementary indexes that can be computed that include a thematic total score for logical memory, visual reproduction, and a number of indexes from a word list learning task. In most instances, summary indexes are constructed in order to label the hypothetical constructs that are thought to underlie performance on the scale in question. As such, the WMS-III summary indexes predict that a particular pattern of covariability exists among the primary and supplementary subtests. For example, general memory is a composite index and implies that the delay recall subtests share a degree of common variability. In other words, there is some common memory construct underlying performance on these subtests. Such models or sets of summary indexes are often called latent variable models as they are defined by the latent constructs that are measured by the subtests (Hayduk, 1987; Joreskog & Sorbom, 1996). Confirmatory factor analysis (CFA) allows the researcher to test a priori hypotheses about the pattern of covariability found in the test data (Joreskog & Sorbom, 1996). Using CFA investigators can empirically address the question of how predictive a particular latent variable model is of actual subtest covariability (Joreskog & Sorbom, 1996). The construct validity of individual test indices can be empirically assessed by testing the extent to which the subtests covary in a manner predicted by the model constructs (Burton, Ryan, Paolo, & Mittenberg, 1994; Nunnally, 1978). CFA is an analytical tool that allows the investigator to explore hypotheses about what constructs the test in question is measuring and provides an empirical basis for clinical interpretation. In CFA the typical procedure is to compare theoretically justified competing models (Gorsuch, 1983; Hayduk, 1987; Hoyle, 1991). This comparison of competing models is the component of CFA that allows the researcher to hypothesis test (Francis, 1988). As hypothesis testing is the primary reason to employ CFA (Nunnally, 1978), failure to evaluate sets of theoretically justified models within relevant samples of subjects limits the validity of the procedure. The predecessor to the WMS-III, the WMS-R (Wechsler, 1987), has been subjected to multiple factor analytic and confirmatory factor analytic studies in both the standardization sample and various clinical samples. On the basis of this analysis, the factors thought to be involved in performance on the WMS-R have included verbal memory, nonverbal memory, attention, immediate, and delay recall (Burton, Mittenberg, & Burton, 1993; Chelune & Bornstein, 1988; Roth, Conboy, Reeder, & Boll, 1990). In contrast, only two confirmatory factor analytic studies have been performed on the WMS-III and both have used the standardization sample. To date, an analysis of a clinical sample has yet to be published. In the analysis presented in the WMS-III technical manual (The Psychological Corporation, 1997) the test publishers reported that they assessed the fit of five models to the pattern of covariability found in the data using three age bands from the standardization sample and the primary subtests. In their analysis the test publishers allowed the error terms of the corre-

3 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) sponding immediate and delayed memory subtests to correlate. The test publisher s models included a one-factor general memory model, a two-factor model that divided the subtests along the working memory versus memory dimensions, and a three-factor model including the factors working memory, immediate memory, and delayed memory. They evaluated a second three-factor model with working memory, visual memory, and auditory memory, and finally a five-factor model that included the dimensions working memory, auditory immediate memory, visual immediate memory, auditory delayed memory, and visual delayed memory. On the basis of their analyses, the test publishers concluded that the five-factor model provided the best fit to the data across the three age bands in the standardization sample. Given that the five-factor model generally corresponded to five of the WMS-III primary indexes (i.e., auditory immediate, visual immediate, auditory delayed, visual delayed, and working memory) the publishers concluded that this finding provided empirical support for the use of these summary measures. Millis, Malina, Bowers, and Ricker (1999) conducted the second CFA of the WMS-III combining the three age bands used by the test publishers into one sample and then replicated the test publisher s analysis. In contrast, to the results reported by the test publishers, Millis et al. (1999) found a three-factor model to be the best fitting. The three-factor model included the dimensions of working memory, visual memory, and auditory memory. Millis et al. (1999) criticized the test publishers as well as their own study for not including supplemental subtests in the analysis. They also noted that their analysis failed to support the immediate and delay memory dimensions in contrast to the analysis conducted by the test publishers using the same normal sample separated into three age bands. Millis et al. (1999) suggested that perhaps the test developers estimated additional parameters in their models that they did not report in the manual. Finally, Millis et al. (1999) emphasized the need for conducting CFA of the WMS-III in clinical samples as different latent variable structures may emerge in these samples that are not apparent in the standardization sample. We agree with Millis et al. (1999) as it is our contention that failure to empirically evaluate the latent variability of the WMS-III using clinical samples calls into question the validity of the confirmatory models that have been portrayed as representing the covariability of the WMS-III subtests. Experience with other Wechsler scales, such as the WAIS-R and WMS-R, clearly suggests that the results of exploratory and CFA often yield results that vary by the demographic and diagnostic characteristics of the samples being analyzed (e.g., Burton et al., 1993; Burton et al., 1994; Leckliter, Matarazzo, & Silverstein, 1986). Additionally, we agree that selected supplemental subtests should be included in the analysis as these subtests are also used to derive summary information from the scale. Finally, as stated by Millis et al. (1999) the issue as to the viability of the immediate and delayed memory indexes needs to be assessed within a clinical sample. This same issue as to whether the immediate versus delayed recall dimensions were viable constructs existed for the WMS-R (e.g., Burton et al., 1993; Roid, Prifitera, & Ledbetter, 1988; Roth et al., 1990) and needs to be empirically addressed for the WMS-III. In the current study we evaluated the construct validity of selected models evaluated by the test publishers and Millis et al. (1999) using a general clinical sample. Structural models were chosen in order to provide a comprehensive test of the latent dimensions that theoretically may underlie performance on the WMS-III. Seven latent variable models were evaluated for

4 632 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) goodness of fit in a clinical sample (n = 281) of adults and then re-evaluated using the WMS-III standardization sample divided into three age bands providing a full crossvalidation of all results. Assessing the seven latent models in both our clinical sample and the standardization samples (n = 1,250) allowed us to empirically determine the extent to which results in the standardization sample generalized to our clinical sample. In addition to the dimensions suggested by the test publishers we tested the viability of a learning factor derived from selected WMS-III word list learning subtests. In reference to the word list learning supplemental subtest, the test publishers make the assertion that high scores indicate efficient learning and immediate recall (The Psychological Corporation, 1997). In the neuroscience literature frontal lesions have been associated with impaired performance on experimental measures of list learning (Mennemeier, Chatterjee, Watson, & Wertman, 1994; Squire, 1987; Squire & Zola-Morgan, 1991; Stuss et al., 1994). The WMS-III word list learning task would appear to be a measure very similar to such tests as the Rey Auditory Verbal Learning Test (RAVLT) or the California Verbal Learning Test (CVLT). Many clinicians routinely use the WMS-R in conjunction with either RAVLT or the CVLT in order to differentiate between memory disorders resulting from frontal and temporal lobe dysfunction and/or in order to discriminate between what some call strategic (learning) versus declarative memory. As a result, we felt it would be clinically useful to empirically test the statistical distinctiveness of the word list learning dimension in comparison to the primary subtests that were reported to be measures of declarative episodic memory by the test publishers (The Psychological Corporation, 1997). 2. Method 2.1. Participants Two hundred and eighty one individuals who had or were suspected of having a neuropathological condition underwent comprehensive neuropsychological evaluations that included the complete WMS-III. Participants were patients at the Veterans Affairs Medical Centers (VAMC) in either Leavenworth, KS or Detroit, MI. As part of the diagnostic work-up, each patient was tested and had their assessment protocols scored by one of two experienced neuropsychologists. The sample was fairly representative of the clinical population found in most VAMC hospitals as it was primarily male (95.8% male, 3.4% female), had a mean age of 51.9 (S.D. = 14.5) ranging from 22 to 87 years, an average education of 12.4 years (S.D. = 2.3), and a mean WAIS-III FSIQ of 88.9 (S.D. = 14.7). Mean performance on the WMS-III immediate memory, general memory, working memory indexes was (S.D. = 16.34), (S.D. = 16.26), and (S.D. = 14.69), respectively. The means and standard deviations of the 14 WMS-III scaled scores in the clinical sample are displayed in Table 1. Sixty-four percent of the sample was Caucasian, 31.4% were African American, and 1.4% were of Hispanic decent. Diagnostic categories included cardiovascular disease (7%), TBI (9%), epilepsy (4%), CNS tumor (.5%), dementia (7%), substance abuse (38%), psychiatric disorder (28%), medical disorders (3%), and normal (4%). The Pearson product-moment intercorrelation matrix used for crossvalidation of results was taken from the WAIS-III WMS-III technical manual (The Psychological

5 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) Table 1 The means and standard deviations of the 14 WMS-III subtests in the clinical sample (n = 281) Subtest Mean S.D. LM I recall Faces I recognition Verbal paired association I recall Family picture I recall Word list I recall Letter/number Spatial span LM II recall Faces II recognition Verbal paired association II recall Auditory recognition delayed Family picture II recall Word list II recall Word list II recognition Corporation, 1997) and included the WMS-III subtest scaled scores of 1,250 normal subjects used in the standardization of the scale divided into three age bands: (n = 400), (n = 400), and (n = 450) (The Psychological Corporation, 1997). Crossvalidation using the standardization sample was performed in order to determine the extent to which results obtained in the VAMC sample were due to idiosyncrasies of that sample. Additionally, as one goal of the current study was to determine whether results in the standardization sample were generalizable to distinct clinical samples it was necessary to conduct the current analysis in both samples Procedure The CFA was performed by using the LISREL 8 software (Joreskog & Sorbom, 1996). The correlations obtained for each pair of WMS-III subtests in the four samples were subjected to CFA by calculating a set of simultaneous structural equations for each of the seven hypothetical models using the Linear Structural Relationship Model. Structural coefficients were estimated from the structural equations using the Maximum Likelihood Fit Function (Hayduk, 1987; Joreskog & Sorbom, 1996). A multivariate probability density formula was used to determine the likelihood that a given set of structural estimates resulted in a difference between the estimated correlation matrix and the actual correlation matrix that was entirely due to chance fluctuation (Hayduk, 1987). The Chi-square statistic divided by its degrees of freedom (d.f.) and the Adjusted Goodness of Fit Index (AGFI) were used to assess model fit. Lower values of the Chi-square/d.f. ratio (χ 2 /d.f.) were assumed to represent a better fit between the hypothesized correlation matrix and the actual correlation matrix. Higher AGFI values were associated with better fitting models. The Parsimonious Normed Fit Index (PNFI) was also used to assess model viability (Netemeyer, Johnston, & Burton, 1990). PNFI values greater than.60 were interpreted as indicating that an increase in the number of freed model parameters resulted in a significant

6 Table 2 The seven hypothetical models LM I Faces I Verbal paired association I Family pictures I Word list I recall Letter/ number Spatial span LM II Faces II Verbal paired association II Auditory recognition delayed Family pictures II Word list II recall Model I General memory Model II General memory Working memory 1 1 Model III General memory Learning Model IV General memory Working memory 1 1 Learning Model V Auditory memory Visual memory Working memory 1 1 Learning Model VI Immediate memory Delayed memory Working memory 1 1 Learning Model VII Immediate auditory 1 1 memory Delayed auditory memory Immediate visual 1 1 memory Delayed visual 1 1 memory Working memory 1 1 Learning Word list II recognition 634 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003)

7 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) increase in the predictive validity of the model over more parsimonious orderings of subtests (Netemeyer et al., 1990). Finally, Chi-squares of difference were computed between selected nested models in order to determine whether freeing additional model parameters significantly increased model fit over simpler models. As we tested increasingly more complex models, it was mandatory that we employed measures to assess the extent to which more complex models provided a justifiable increase in model fit over more parsimonious models. The seven models to be tested were derived on the basis of previous research with the WMS-R and WMS-III, the CFA of the WMS-III presented by the test publishers, the analysis of Millis et al. (1999), and selected theoretical models of memory that the WMS-III was proposed to measure (e.g., the learning dimension). Model I was a one-factor general memory model that corresponded to the first model assessed by the test publishers. Likewise, Model II contained the factors working memory and general memory corresponding to the test publisher s second model. Model III was a two-factor model that assessed the fit of general memory and learning factors. In Model III, the learning factor was composed of the word list learning I recall, word list learning II recall, and word list learning II recognition subtests. Model IV was a three-factor model that included general memory, working memory, and learning factors. Model V subdivided the general memory factor into auditory memory and visual memory with the working memory and learning factors. Model VI divided the general memory factor into immediate memory and delay memory again paired with the working memory and learning factors. Finally, Model VII included the factors immediate auditory memory, delay auditory memory, immediate visual memory, delay visual memory, working memory, and learning. Model VII represented the test publisher s most complex and proposed best fitting model with the addition of the learning factor. The error terms from the immediate and delayed recall versions of each subtest (e.g., logical memory, facial recognition, verbal paired associates, family pictures, and word list learning) were allowed to correlate, which duplicated the test publisher s analysis with the addition of the word list learning subtests. Additionally, the error terms for the immediate and delayed recall versions of verbal paired associates and word list learning were allowed to correlate as well as auditory recognition delayed and word list II recognition in order to model declarative versus strategic memory factors. The seven models are summarized in Table Results The results of the maximum-likelihood CFA of the general clinical sample are summarized in Table 3. Review of Table 3 reveals that Model V displayed the lowest Chi-square/d.f. ratio (χ 2 /d.f. = 2.44), an AGFI of.88, and a PNFI of.66. The results of the analysis of the standardization sample are presented in Table 4and are generally equivalent to the results of analysis in the clinical sample. Model V displayed the lowest Chi-square/d.f. ratio across two of the three age ranges of the standardization sample (age 16 29: χ 2 /d.f. = 2.40 and age 65 89: χ 2 /d.f. = 2.72), AGFI s of.91, and PNFI s of.66. For the year-olds standardization sample Model VII displayed the lowest Chi-square/d.f. ratio (χ 2 /d.f. = 2.60) with Model V displaying the next lowest Chi-square/d.f. ratio (χ 2 /d.f. = 2.69). Further, examination of AGFI s revealed that Model VII displayed a minimally higher AGFI compared

8 636 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) Table 3 Summary of goodness-of-fit statistics for the clinical sample (n = 281) Models χ 2 d.f. χ 2 /d.f. AGFI PNFI I II III IV V VI VII to Model V (Model V AGFI =.90 and Model VII AGFI =.91) in the year-old sample. In contrast, examination of the PNFIs revealed that Model VII fell below the commonly employed criteria point of.60 (PNFI =.57) in all four samples. This suggested that the increase in the number of freed model parameters in Model VII did not result in a significant increase in predictive validity over more parsimonious orderings of subtests in any of the four samples. Table 4 Summary of goodness-of-fit statistics for the standardization samples Models χ 2 d.f. χ 2 /d.f. AGFI PNFI Ages (n = 400) I II III IV V VI VII Ages (n = 400) I II III IV V VI VII Ages (n = 450) I II III IV V VI VII

9 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) Table 5 Selected nested comparisons in the clinical and standardization samples Standardization sample Clinical sample Ages Ages Ages χ 2 d.f. P χ 2 d.f. P χ 2 d.f. P χ 2 d.f. P Comparisons I versus II < < < <.001 I versus III < < < <.001 II versus IV < < < <.001 III versus IV < < < <.001 IV versus V < < < <.001 IV versus. VI > > > >.05 V versus VII > > < <.02 As an additional check of model viability nested comparisons were performed in both the clinical and standardization samples between selected models. Chi-squares of difference were computed to determine whether selected models resulted in a statistically improved model fit compared to more parsimonious models. Inspection of Table 5 reveals that the two-factor models II (general memory and working memory) and III (general memory and learning) fit significantly better than the one factor general memory model across all four samples. Likewise, the three-factor Model IV (general memory, working memory, and learning) fit better than both two-factor models (II and III) across all four samples. Model V provided a statistically significant improvement in goodness of fit over the three-factor Model IV in all four samples. In contrast, Model VI failed to provide a statistically significant increase in model fit compared to the more parsimonious Model IV across all four samples. A comparison of Model V versus Model VII revealed that Model VII failed to provide a significant increase in model fit over the more parsimonious Model V in the clinical and the year-old standardization sample. In contrast, Model VII did result in a significant improvement in model fit compared to Model V in the and year-olds standardization samples. Consensus of results across all four samples supports Model V as the best fitting model of those models tested in the current analysis. Model V had the lowest Chi-square/d.f. ratio in three out of four samples, the highest or an equivalent AGFI in three out of four samples, and a PNFI suggesting that the increase in model complexity was justified by an increase in model fit across all four samples. Nested comparisons using Chi-square difference tests also supported the fit of Model V over more parsimonious ordering of subtests in all four samples and over more complex models in two out four samples. In the year-olds standardization sample Model VII had a lower Chi-square/d.f. ratio and a minimally higher AGFI compared to Model V. However, across all four samples Model VII failed to meet criteria in regards to the PNFI suggesting that the increase in model parameters represented by Model VII did not result in a significant improvement in model fit compared to more parsimonious models in any of the samples. The standardized oblique structural coefficients computed from the clinical sample for Model V are displayed in Table 6.

10 638 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) Table 6 Completely standardized structural coefficients for Model V in the clinical sample Auditory memory Visual memory Working memory Learning LM I recall Faces I recognition Verbal paired association I recall Family picture I recall Word list I recall Letter/number Spatial span LM II recall Faces II recognition Verbal paired association II recall Auditory recognition delayed Family picture II recall Word list II recall Word List II recognition Discussion CFA provided support for the presence of four underlying latent variables in the WMS-III performance of our general clinical sample, and findings were crossvalidated in all three age ranges of the standardization sample. Results suggested that the best predictor of intersubtest variability, among the set of models evaluated in the current analysis across both the clinical and standardization samples was Model V. Model V was significantly more accurate in explaining the intersubtest variability of the WMS-III and generally provided a better fit to the data across all four samples compared to both more parsimonious and more complex models. Model V contained the latent variables auditory memory, visual memory, working memory, and learning. Our findings are consistent with those of Millis, Malina, Bowers, and Ricker (1999) who performed a CFA of the WMS-III in the standardization sample. They assessed the fit of five-factor models that were nearly identical to the models evaluated in the current analysis (Millis et al., 1999). Millis et al., found that a three-factor model containing the latent variables auditory memory, visual memory, and working memory fit the best in the standardization (n = 2,450) sample. The results of our analysis of the WMS-III in our clinical and the standardization samples are nearly identical to the results of their CFA (e.g., Millis et al., 1999) with exception that they did not assess a learning factor. In our analysis we also found that the immediate versus delayed memory dimensions were not empirically supported. Thus, Millis et al. s (1999) hypothesis that these dimensions may emerge when analyzed in a clinical sample was not supported in our study. In contrast, the results of our analysis in comparison to the CFA presented in the WMS-III manual, supports a different interpretation of performance on the test. The test publishers reported that their five-factor model containing the dimensions of working memory, visual immediate memory, auditory immediate memory, visual delayed memory, and auditory delayed memory fit the best in the standardization samples. Our analysis failed to find consis-

11 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) tent support for this model and for the immediate versus delayed memory dimensions. One potential reason for this discrepancy in findings is that the test publishers did not evaluate latent variable models that included supplementary subtests despite promoting the use of these subtests for clinical interpretation. As Millis et al. (1999) points out it might be possible to find... additional latent constructs among the supplemental measures. Another possible reason, as reported by Millis et al. (1999), is that the test publishers may have estimated additional parameters in their models that they did not report in the manual in order to achieve greater goodness of fit. We completely agree with Millis et al. (1999) that disclosure of all model parameters assessed in the standardization of the scale will allow for the possibility of replication and will provide a means for empirically assessing the clinical utility of the indexes offered by the test publishers. Failure to do so hampers the empirical validation of the scale. In addition to the question of how well a model fits empirically, it is equally important to address the issue of whether that model is theoretically valid (Joreskog & Sorbom, 1996). One of the characteristics of CFA that makes it appealing from a methodological perspective is that hypotheses in the form of models to be tested are constructed a priori. Models to be tested are chosen on the basis of previous research and on the basis of the theoretical properties of the processes that one is attempting to measure. One of the characteristics of Model V that makes it more compelling is its correspondence to accepted models of memory function. In the neuroscience literature frontal lesions have been associated with impaired performance on experimental measures of list learning and attention/concentration, while story recall, paired-associate learning, and figural reproduction have generally been associated with temporal lobe pathology (Mennemeier et al., 1994; Squire, 1987; Squire & Zola-Morgan, 1991; Stuss et al., 1994). Theoretically, the frontal lobes have been assigned the role of a central system structure that operates strategically on the input to the medial temporal lobes/hippocampus (Moscovitch, 1994). Moscovitch (1994) has argued that the prefrontal cortex serves to conceptually and meaningfully organize information prior to the encoding process conducted by the medial temporal lobes and hippocampus. Along these same lines, other investigators have suggested that the frontal lobes subserve the ability to conceptually organize information while strategically coordinating the act of learning in a way that facilitates declarative memory as subserved by the temporal lobes (Gershberg & Shimamura, 1995; Shimamura, Janowsky, & Squire, 1991; Squire, 1987; Stuss et al., 1994). It is our proposition that the list learning tasks from the WMS-III provide a measure of the individual s ability to conceptually organize information in a manner that facilitates their auditory and visual declarative memory. This is consistent with the tests publisher s definition of the list learning tasks as they describe the task as a measure of efficient learning and immediate recall (The Psychological Corporation, 1997). We agree with the WMS-III publishers and also view the working memory factor as a measure of active mental processing consistent with the previous definitions of the Attention/Concentration Index from the WMS-R. Finally, we view the auditory and visual memory Factors as measures of auditory declarative memory primarily subserved by the language dominant hemisphere and visual declarative memory primarily subserved by the nondominant hemisphere. Our four-factor model (Model V) was empirically more accurate in predicting the pattern of latent variability among the subtests compared to the test publisher s models. This suggests that the present model provides a more valid way of representing the dimensions of memory

12 640 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) as measured by the WMS-III. As such, the implication is that when summarizing the results of the WMS-III for an individual examinee, it might be appropriate to combine the subtests in a manner consistent with our four-factor model. However, there still remains the question of clinical utility. In this regard, future research needs to address the empirical question as to whether our four factors can be used to differentiate distinct clinical samples on the basis of profile analysis. While the four-factor model may be the best fitting in the set of models we tested, the final measure of its utility must lie in assessing its usefulness in making clinical decisions. Finally, the fit of our models should be assessed in other clinical samples to provide a further crossvalidation of the current findings. The clinical utility of the WMS-R has been significantly enhanced as the result of clinical validation. It is our contention that the usefulness of the WMS-III can also be enhanced through such research. References Burton, D. B., Mittenberg, W., & Burton, C. A. (1993). Confirmatory factor analysis of the WMS-R standardization sample. Archives of Clinical Neuropsychology, 8(6), Burton, D. B., Ryan, J. J., Paola, A., & Mittenberg, W. (1994). Structural equation analysis of the WAIS-R in a normal elderly sample. Psychological Assessment, 6(4), Chelune, G. J., & Bornstein, R. A. (1988). WMS-R patterns among patients with unilateral brain lesions. Special issue: Initial validity studies of the new Wechsler Memory Scale-Revised. Clinical Neuropsychologist, 2(2), Francis, D. J. (1988). An introduction to structural equation models. Journal of Clinical and Experimental Neuropsychology, 10(5), Gershberg, F. B., & Shimamura, A. P. (1995). Impaired use of organizational strategies in free recall following frontal lobe damage. Neuropsychologia, 33(10), Gorsuch, R. L. (1983). Factor analysis. New Jersey: Lawrence Erlbaum Associates. Hayduk, L. A. (1987). Structural equation modeling with LISREL. Baltimore: John Hopkins Press. Hoyle, R. H. (1991). Evaluating measurement models in clinical research: Covariance structure analysis of latent variables models of self-conception. Journal of Consulting and Clinical Psychology, 59(1), Joreskog, K. G., & Sorbom, D. (1996). LISREL 8: User s reference guide. Chicago: Scientific Software International, Inc. Leckliter, I. N., Matarazzo, J. D., & Silverstein, A. B. (1986). A literature review of the factor analytic studies of the WAIS-R. Journal of Clinical Psychology, 42, Mennemeier, M. S., Chatterjee, A., Watson, R. T., & Wertman, E. (1994). Contributions of the parietal and frontal lobes to sustained attention and habituation. Neuropsychologia, 32(6), Millis, S. R., Malina, A. C., Bowers, D. A., & Ricker, J. H. (1999). Confirmatory factor analysis of the Wechsler Memory Scale-III. Journal of Clinical and Experimental Neuropsychology, 21(1), Moscovitch, M. (1994). Cognitive resources and dual-task interference effects at retrieval in normal people: The role of the frontal lobes and medial temporal cortex. Neuropsychology, 8(4), Netemeyer, R. G., Johnston, M. W., & Burton, S. (1990). Analysis of role conflict and role ambiguity in a structural equations framework. Journal of Applied Psychology, 75, Nunnally, J. C. (1978). Psychometric theory. New York: McGraw-Hill. Roid, G. H., Prifitera, A., & Ledbetter, M. (1988). Confirmatory analysis of the factor structure of the Wechsler Memory Scale-Revised. Clinical Neuropsychologist, 2(2), Roth, D. L., Conboy, T. J., Reeder, K. P., & Boll, T. J. (1990). Confirmatory factor analysis of the Wechsler Memory Scale-Revised in a sample of head-injured patients. Journal of Clinical and Experimental Neuropsychology, 12(6),

13 D. Bradley Burton et al. / Archives of Clinical Neuropsychology 18 (2003) Shimamura, A. P., Janowsky, J. S., & Squire, L. R. (1991). What is the role of frontal lobe damage in memory disorders? In H. S. Levin, H. M. Eisenberg, & A. L. Benton (Eds.), Frontal lobe function and dysfunction (pp ). New York: Oxford University Press. Stuss, D. T., Alexander, M. P., Palumbo, C. L., Buckle, L., Sayer, L., & Pogue, J. (1994). Organizational strategies in patients with unilateral or bilateral frontal lobe injury in word list learning tasks. Neuropsychology, 8(3), Squire, L. R. (1987). Memory and brain. New York: Oxford University Press. Squire, L. R., & Zola-Morgan, S. (1991). The medial temporal lobe memory system. Science, 253(5026), The Psychological Corporation. (1997). WAIS-III WMS-III technical manual. San Antonio, TX: Harcourt Brace. Wechsler, D. (1987). Wechsler Memory Scale-Revised. San Antonio, TX: The Psychological Corporation.