Measurement of premorbid intellectual ability following brain injury

Pergamon

Archiv~of ClinicalNeuropsychology,Vol. I I. No. 6, pp. 491-501. 1996 Copyright© 1996NationalAcademyof Neuropsychology Printedin the USA.All rights re,fred 0887-6177/96$15.00+ .00

SSD! 0887-6177(95)00051-8

Measurement of Premorbid Intellectual Ability Following Brain Injury Stephanie A. Perez Mediplex-Rehab-Cortland

Robert S. Schlottmann Oklahoma State University

Joan A. Holloway University of Oklahoma Health Sciences (:enter

Mickey S. Ozolins NeuroBehavioral Institute

Prediction of premorbid intellectual ability in brain-injured patients was investigated using two sets of regression equations and the Intellectual Correlates Scale (ICS). Eighty subjects completed the WAIS-R and the ICS. The four subject groups included a control group and right-hemisphere, lefthemisphere, and diffuse brain-injured groups. As expected, brain-injured groups obtained lower IQs than controls. Also, estimated IQs approximated obtained IQs for controls, while overestimating IQs for brain-injured groups. Support was provided for the continued use of the Barona, Reynolds, and Chastain (1984) and the Barona and Chastain (1986) regression equations as measures of prenmrbid intellectual functioning. Previous findings (Schlottmann & Johnsen, 1991), suggesting the ICS may also serve as a measure of premorbid intellectual functioning, were not replicated.

In working with brain-injured patients, knowledge of premorbid intellectual functioning can assist the clinician in assessing the extent of impairment, formulating a maximally effective treatment plan, identifying potential difficulties in readjustment, and dealing with legal concerns. However, it is often difficult to obtain accurate information regarding premorbid abilities, as previous psychological testing is rarely available (Chelune, Ferguson, & Moehle, The authors would like to thank the administrators and staffs at NeuroBehavioral Institute, St. Anthony's Hospital, and South Community Hospital for their assistance in making this study possible. Address correspondence to: Stephanie A. Perez, Mediplex-Rehab-Cortland, 24 Groton Avenue, Cortland, NY 13045. 491

492

S. A. Perez et al.

1986). Two methods of attempting to obtain such information, which will be investigated in this article, are (a) the use of regression equations containing demographic information and (b) the assessment of attitudes, interests, and beliefs that correlate with intelligence. It is noted that other methods for assessing premorbid abilities have been developed, for example, performance on the Vocabulary subtest (Lezak, 1983) and use of Hold and Don't Hold subtests (Hunt, 1949). These other methods are not further addressed in this article, as a patient's comparative ability on different tasks is dependent upon the location and severity of brain injury (Klesges & Troster, 1987) and, thus, could potentially lower the premorbid estimates. The use of demographic regression equations and/or the attitudes questionnaire in estimating premorbid abilities would be expected to not be impacted by the effects of the brain injury. Barona, Reynolds, and Chastain (1984) developed a set of regression equations (B84) on the 1981 WAIS-R standardization sample. The B84 uses the demographic variables of age, .sex, race, education, occupation, region of the country, and urban or rural residence to predict intellectual ability, as measured by the WAIS-R. Barona et al. (1984) suggested that these equations could estimate premorbid functioning in head-injured patients. The B84 equations correlated .60 with WAIS-R FSIQ, .62 with VIQ, and .49 with PIQ. Eppinger, Craig, Adams, and Parsons (1987) successfully cross-validated the B84 using 163 subjects who were either neurologically-impaired or normal. Controls with a history of psychosis, substance dependence, or congenital mental retardation were not used, although other psychological disturbances were allowed. Eppinger et al. (1987) found that differences between WAIS-R and B84 IQs were higher for the impaired group and that the B84 correlated .78 with VIQ, .60 with PIQ, and .76 with FSIQ. The use of the B84 allowed correct classification of the majority of subjects, but not at a level better than use of the WAIS-R alone. Eppinger et al. (1987) pointed out several limitations of the B84 equations, namely that their accuracy decreases when obtained IQs are below 69 or above 120, that the occupational and educa,.ional categories often are not specific enough to account for individual cases, that motivation may affect WAIS-R IQ but cannot be accounted for by the B84, and that the B84 tends to overestimate WAIS-R IQs, regardless of group membership. They suggested that the tendency to overestimate WAIS-R IQs in the control group could be due to psychological disturbances, and they discussed the use of difference scores with the B84 in individual cases. Bar0na and Chastain (1986) recalculated the B84 equations including only Black or White subjects, 20 years of age and above, in order to increase the accuracy of prediction with respect to Black and White adults. The refined equations (B86) correlated .65 with WAIS-R FSIQ, .68 with VIQ, and .53 with PIQ. Both the B84 and B86 use the same variables, but these variables are weighted differently in each set of equations with respect to estimating WAIS-R VIQ, PIQ, and FSIQ. Both sets of equations have large standard errors of estimate, which decreases their helpfulness in individual cases. An alternative approach is that of using the Intellectual Correlates Scale (ICS), which is based on interests, attitudes, and beliefs found to correlate with intellectual functioning (Gentry, 1982; Schlottmann & Johnsen, 1991). This type of approach theoretically allows consideration of individual differences to a greater extent than the regression equations; however, it does not take demographic variables into account. Gentry (1982) initially developed items for inclusion in the ICS, based upon their correlations with the Shipley. Gentry (1982) administered the items and the Shipley to 100 college undergraduates, retaining those items correlating with performance on the Shipley. He cross-validated the resulting ICS scale on 30 undergraduates who took the ICS, Shipley, and a short version of the WAIS. A correlation of .66 was found between ICS and Shipley scores. Estimated scores were converted to Shipley scores and compared with the WAIS

Preraorbid Intellectual Abilities

493

IQs. Gentry (1982) found that WAIS IQs were significantly underestimated for males, but that no significant mean differences occurred for either females or for the total group, suggesting accuracy in estimating these IQs. Since Gentry (1982) found higher correlations for verbal than for performance scores, he suggested developing separate scales for estimating VIQ and PIQ. Gentry (1982) speculated that the ICS could be useful in estimating premorbid intellectual ability in cases of acute brain injury, but not in chronic cases, as an individual's interests and attitudes could change over time as he or she adapted to the injury. The ICS was further developed by Johnsen (1987). He used 33 adults below 60 years of age to investigate the correlation between ICS items and WAIS-R IQs, retaining the 71 items with the strongest correlations. He also devised equations to predict WAIS-R VIQ, PIQ, and FSIQ and cross-validated the scale, administering the WAIS-R to 64 brain-injured and control subjects, matched on age and education. B84 estimates were also calculated. Johnsen (1987) concluded that "the three ICS-based IQ estimates accounted for a greater percentage of the variance in obtained IQs than did the Barona-based estimates" (p. 44). Standard errors of estimate for the ICS were 9.80 for VIQ, 10.20 for PIQ, and 9.22 for FSIQ. Schlottmann and Johnsen (1991) reanalyzed the Johnsen (1987) data to include the use of the B86 equations. They dropped out three of the control subjects who were not appropriate for inclusion when using the B86. The ICS was found to correlate .57 with WAIS-R VIQ, .54 with P1Q, and .65 with FSIQ. The present research involved an attempt to cross-validate and compare the ICS, B84, and B86. Also, the effects of lateralization upon the estimation procedures were investigated. Previous research has indicated that individuals with right hemisphere lateralized injury tend to have greater Verbal than Performance IQs, while individuals with left hemisphere lateralized injuries tend to have greater Performance than Verbal scores, although this latter finding is reported to occur much less consistently (Kaufman, 1990, p. 299). It was hypothesized that: 1. Estimated IQs for controls would reflect current intellectual functioning, resulting in no differences between estimates and WAIS-R IQs. 2. For each brain-injured group, the estimated IQs were expected to be higher than WAIS-R IQs, because the IQs, but not the estimates, should be affected by intellectual impairment resulting from brain injury. 3. Individuals with left-hemisphere lateralized injury were expected to score lower on Verbal IQ than those in the right hemisphere and control groups, whereas individuals with righthemisphere lateralized injury were expected to score lower on Performance IQ than those in the left hemisphere and control groups. Lateralized effects were not expected in the diffuse group as they should do poorly on both Verbal and Performance IQ. 4. In the control group, IQ estimates were expected to correlate positively with WAIS-R IQs, reflecting the validity of the estimators.

METHOD Subjects

Brain-injured subjects were recruited from two rehabilitation hospital units and from an outpatient head injury clinic in Oklahoma City. Controls were either referred into the study by other subjects and persons familiar with the research or recruited through hospital records following hip replacement surgery. Subjects were required to (a) be between the ages of 20 and 74, (b) be either Black or White, and (c) be sufficiently high functioning to be administered the WAIS-R and the ICS. Subjects were selected for, and assigned to, one

494

S. A. Perez et al.

of four groups based on neurological and/or neurosurgical records contained in their medical charts. When available, CAT scan and MRI data were used to confirm group classification. Neuropsychological test data was not used in assigning subjects to groups. The four groups were (a) left-hemisphere brain-injury, (b) right-hemisphere brain injury, (c) diffuse brain-injury, and (d) a control group with no reported history of brain injury or psychiatric impairment. Subjects were sought who met the above criteria and were only assigned to a group when they clearly fit all selection criteria. Group size was limited to 20 subjects per group in order to have equal ns because ANOVAs with different numbers of subjects in the cells have been found to be nonrobust (Milligan, Wong, & Thompson, 1987). Ninety-eight subjects were recruited. Of these, 17 were disqualified due to either reports of a previous brain injury, an inability to complete the testing, or a failure to meet the general requirements outlined above. One subject was disqualified due to alcohol consumption. Thus, 80 subjects completed participation in the research. Each of the four groups consisted of 20 subjects. One-half of the subjects in each of the brain-injured groups had sustained brain injury less than 4 months prior to participation in the study (acute condition) while the other one-half of the subjects had sustained injury greater than 6 months prior to participation (chronic condition). Each control referred by a member of an acute brain-injury group was assigned to the acute control group, whereas each control referred by a member of a chronic brain-injury group was assigned to the chronic control group. Remaining controls were randomly assigned using a random numbers table. The use of two control groups avoided the difficulties of performing a complex factorial design with a single control group. Also, initially there were concerns that the acute and chronic groups might differ in unknown ways because they tended to come from different sources (e.g., most of the acute subjects came from a single hospital). Due to the clinical nature of the brain-injured sample and the inability to experimentally assign subjects to groups, demographic distributions also varied between the left, right, diffuse, and control groups. For example, subjects suffering from diffuse injuries such as motor-vehicle accidents and work injuries, tended to be younger than subjects suffering from lateralized injuries, such as strokes. Table 1 provides descriptive data for each group. Materials

The materials administered to each subject were the WAIS-R, the ICS, a demographic questionnaire, and a medical history questionnaire. The ICS is a 71-item self-rating scale based on interests, attitudes, and beliefs found to correlate with WAIS-R IQ (Johnsen, 1987). The subject responds to each item by indicating whether he or she strongly agrees, agrees, disagrees, or strongly disagrees with the item. In a previous study (Schlottmann & Johnsen, 1991), the ICS verbal, performance, and full scale scores were found to correlate .57, .54, and .65 with WAIS-R Verbal, Performance, and Full Scale IQ. The ICS scores were found to have Standard Error of Estimates of 9.41 to 9.89. Procedure

Subjects were assigned to the appropriate group by the researcher after review of their medical charts. Historical information and data obtained from the Medical History Questionnaire were used to screen all subjects for a significant history of brain-injury and psychiatric disorders. Subjects who reported a history of brain injury or significant psychiatric disorders were not used in the control group. Subjects with evidence of a history of confounding brain injury, thus making group assignment difficult, were not used in forming the brain-injured groups. Due to the acute nature of the injury and possible severity of resulting effects, the

Premorbid Intellectual Abilities

495

TABLE 1 Demographic Characteristics of Subject Groups Group Variable

Right

Left

Diffuse

Controls

Mean age

62.15 12.82 12.95 4.67 11/9 18/2 16/4 15/1 4/0 4 9 3 0 2 2

59.30 12.63 II .45 2.93 1I/9 17/3 1218 18/1 I/0 2 6 4 2 5 1

37.05 17.15 11.85 1.87 18/2 20/0 14/6 1910 0/I 2 3 11 2 2 ()

47.10 16.34 14.2 2.22 5/I 5 29/I 15/5 15/0 5/0 10 7 2 1 () 0

15/3 I 0/0 0 0 I

16/2 I 0/I 0 0 0

1/0 0 14/0 2 2 1

0/0 0 0/0 0 0 0

SD Mean education

SD Male/Female White/Black Urban/Rural South/Northeast Northcentral/West Professional Managerial Skilled Unemployed Semi-skilled Unskilled Cerebrovascular accident/aneurysm Meningioma Closed/Open head injury Encephalitis Anoxia Hematoma

physician at the inpatient units made a determination regarding the competency of potential subjects to make informed decisions about their participation. Only those subjects deemed competent to give consent, or those who gave their consent along with the consent of the responsible family member, were allowed to participate. The order of test administration was: Demographic Questionnaire, Medical History Questionnaire, WAIS-R, and ICS. Verbal, Performance, and Full Scale IQs were obtained from the WAIS-R. Estimates of Verbal, Performance, and Full Scale IQs were obtained from the ICS and the B84 and B86 equations. Subjects were informed that the purpose of the research was to investigate the measurement of intellectual ability in individuals with and without brain injury. They also were told that they would be administered an intelligence test and a questionnaire about their attitudes, interests, and beliefs. It was explained to all subjects that they were free to withdraw from the study at any time and that all information would be kept confidential. After the subject completed his or her participation in the study, brief feedback, supervised by a clinical neuropsychologist, was given. Feedback involved a review of each subject's general performance and areas of strengths and weaknesses. When subjects were referred from an inpatient or outpatient unit, a copy of the test data was provided to the inpatient or outpatient unit when appropriate.

RESULTS A 4 x 2 ANOVA was performed on Groups (Right, Left, Diffuse, and Control) x Duration (Acute, Chronic) on age and education. A significant main effect was found between groups for both age, F(3, 72) = I 1.89, p < .0001 and education, F(3, 72) = 3.19, p < .05. Pairwise comparisons using Tukey's Studentized Range showed that right brain-injured subjects

496

S. A. Perez et al.

(M = 62.15) were significantly older than diffuse subjects (M = 37.05) and that controls (M = 14.20) were significantly more educated than left brain-injured subjects (M = 11.45). Other groups did not differ on age and education. Review of the demographic characteristics of subject groups also revealed large differences in gender between groups. Thus, a 4 x 2 x 4 multivariate analysis of covariance was performed on the verbal, performance, and full scale scores using gender (coded as 1 or 0), age, and education as covariates. The independent variables were Group (Left, Right, Diffuse, and Control), Duration (Acute or Chronic), and Procedure used to derive scores (ICS, B84, B86, and WAIS-R). Using the Wilks' lambda criterion, a significant Group x Procedure interaction was found,/7(27, 626) = 8.76, p < .0001. Thus, univariate analyses were carried ouL and a significant Group x Procedure interaction was found on all three dependent variables, F(9, 216) = 8.09, 14.49, and 8.34 for verbal, performance, and full scale scores, respectively, p < .0001. Mean scores, unadjusted for covariate effects, are shown in Table 2. Using the SAS PDIFF option for planned comparisons, it was found that, for the control group, there were no significant (p > .05) differences between IQs (Verbal, Performance, and Full Scale) and any of the respective estimated scores. On the other hand, each of the estimated scores were significantly (p < .001) higher than their respective IQs in the left hemisphere, right hemisphere, and diffuse brain-damage groups. These tests were done on the unadjusted means because they were within-group comparisons. The results were as predicted in Hypotheses 1 and 2. To confirm that the brain-damaged groups were intellectually impaired, the SAS PDIFF option was used to compare Groups on WAIS-R IQs adjusted for differences in age, education, and sex. The control group's Verbal IQ was significantly (p < .001) higher than that of each of the groups with brain damage (see Table 3). Also, subjects with left hemisphere brain damage scored significantly (p < .003) lower than those with right hemisphere damage. For Performance IQ, the control group's mean score was significantly (p < .0001) higher than those of each of the groups with brain damage, and subjects with right hemisphere damage scored significantly (p < .002) lower than those with left hemisphere damage. Finally, for Full Scale IQ, the control group scored significantly (p < .0001) higher than each of the groups with brain damage, but there were no significant differences in Full

TABLE 2

Means Verbal, Performance, and Full Scale Scores on the ICS, 13/14, B86, and WAIS-R by Group Score Group

Procedure

Verbal

Performance

Full Scale

Control

ICS B84 B86 WAIS-R ICS B84 B86 WAIS-R ICS B 84 B86 WAIS-R ICS B84 B 86 WAIS-R

106.7 110.5 108.9 108.3 109. I 105.4 106.4 96.8 108.0 100.0 100.0 83.2 109.0 103.2 100.4 86.6

I 10.3 107.4 107.1 108.4 107.8 103.0 105.0 77.8 104.2 99.0 100.0 90.0 106. I 101.8 101.0 81.5

108. I 109.8 109.1 109.2 107.3 104.6 106.1 88.0 104.0 99.4 100.1 85.4 105.4 102.6 100.7 83.4

Right

Left

Di ffu~

Premorbid Intellectual Abilities

497

TABLE 3

Mean IQs for Each Group Adjusted for Differences in Age, Sex, and Education

IQ Group

Verbal

Performance

Full Scale

Control Right Left Diffuse

107.9 93.9 82.6 90.5

108.4 77.0 89.1 82.8

108.9 85.7 84.8 86.7

Scale IQ between any of the groups with brain damage. These results were consistent with Hypothesis 3. They simply confirm that the brain-damaged groups scored lower on the WAIS-R than the controls even when age, sex, and education were taken into account, a necessary finding if intellectual loss is to be investigated. Because there were no predictions about differences between estimation procedures, Tukey's procedure was used to investigate pairwise comparisons other than those hypothesized. Results indicated that the ICS, B84, and B86 estimated IQs did not differ significantly from each other for the control, left brain-injured, or right brain-injured groups. In the diffuse group, however, the ICS estimates were significantly higher than the B84 estimates on VIQ and higher than the B86 estimates on VIQ, PIQ, and FSIQ (see Table 2). Pearson correlation coefficients were used to investigate the relationships between the estimators and WAIS-R (Hypothesis 4). Only control group data were used in these computations because it is solely in this group that estimates would be expected to predict current WAIS-R IQs. Data displayed in Table 4 indicate that the estimated IQs from both the B84 and B86 correlated significantly with WAIS-R IQs. ICS scores, however, were not significantly correlated with any WAIS-R IQs. Thus, although the sample mean IQ produced by the estimator was the same as the actual IQ sample mean, there was no correlation between ICS and IQ. Since groups differed on gender, age, and education, regression equations were generated by using the SAS PROC REG procedure to see if the ICS could be improved by taking these variables into account. Multiple correlations, using the three predictors of age, gender, and education were .449 for VIQ, .252 for PIQ, and .427 for FSIQ. Adding in the ICS, the multiple correlations with four predictors were .633 for VIQ, .253 for PIQ, and .487 for FSIQ. Using squared semipartial correlations, it was found that the ICS did not significantly improve IQ estimation beyond that attained by the use of demographic factors for either VIQ, F(3, 16) = 2.68, p > .05, PIQ, F(3, 16) = .008, p > .05, or FSIQ, F(3, 16) = .0396, p > .05. A classification analysis (Huberty, 1984) using the B84 and B86 equations to classify left, right, and diffuse brain-injured, and control subjects was conducted. This analysis is TABLE 4 Correlations of WAIS-R IQs With Estimated IQs on the B84,1386, and ICS in the Control Group Estimation Proce,dure WAIS-R

B84

B86

ICS

VIQ PIQ FSIQ

.52* .27 .48*

.52* .21 .52*

.OI .02 .17

*p < .05

498

S. A. Perez et al.

particularly useful when analyzing actual versus predicted classifications using frequency data. Table 5 fists the number and percentage of subjects classified into each group, along with the z statistics for frequency of hits in each group. The proportional chance criterion for estimating prior probabilities was used. ICS estimates were not used in this analysis because they did not correlate with WAIS-R IQs in the control group. Subjects were classified as having right-hemisphere injury when their WAIS-R PIQ was more than one Standard Error of Estimate below their estimated PIQ, as having left-hemisphere injury when their WAIS-R VIQ was more than one Standard Error of Estimate below their estimated VIQ, and as having diffuse brain injury when both their WAIS-R VIQ and PIQ were more than one Standard Error of Estimate below their corresponding estimated VIQ and PIQ. All other subjects were classified as controls. The improvement over chance classification for the control group was 86.66% using the B84 and 93.33% using the B86. For the diffuse group, the improvement over chance classification was 40% using the B84 and 33.33% using the B86. Although both the B84 and B86 classified controls and diffuse brain-injured subjects significantly better than chance, neither classified the right or left hemisphere-injured subjects at a better than chance level. Because of the small sample size, a second classification analysis was conducted in which subjects were categorized as either brain-injured or control. Subjects were predicted to be brain injured if their estimated VIQ, PIQ, or FSIQ was more than one Standard Error of Estimate greater than their corresponding WAIS-R IQ. Otherwise, subjects were classified as controls. The B84 correctly classified 76.67% of the brain-injured group and 90% of the control group. The B86 correctly classified 80% of the brain-injured group and 95% of the control group. Ratios of percentages of false negatives to false positives were 23.33:10.00 using the B84 and 20.00:5.00 using the B86. Using Dunn's procedure, the critical value for z, one-tailed, is 1.96. Neither the B84 (z = .2981, p > .05) nor the B86 (z = .8944, p > .05 classified brain-injured subjects any better than chance, even though both classified the majority of subjects correctly. However, the control group was classified by the B84 with an 86.66% improvement over chance (z = 6.71, p < .05) and by the B86 with a 93.33% improvement over chance (z = 7.23, p < .05).

DISCUSSION As expected, the ICS, B84, and B86 were found to approximate current functioning in the control group, while overestimating current functioning in each brain-injured group. TABLE 5 Number and Percentage of Subjects Classified in Control, Diffuse, Right, or Left Brain-Damaged Groups, and z-Statistics Using Either the B84 or B86 and WAIS-R Discrepancy Scores Predicted Group Me~ure B84

B86

Actual Group

Control

Diffuse

Left

Right

z

Control Diffuse Left Right Control Diffuse Left Right

18(90%) 1 (5%) 10(50%) 3(15%) 19(95%) I (5%) 8(40%) 3(15%)

I (5%) 11(55%) 8(40%) 7(35%) I (5%) 10(50%) 8(40%) 8(40%)

I (5%) 3(15%) I (5%) I (5%) 0 (0%) 4(20%) 2(10%) 0 (0%)

0 (0%) 5(25%) I (5%) 9(45%) 0 (0%) 5(25%) 2(10%) 9(45%)

6.71" 3.10" -2.07 2.07 7.23* 2.58* -I.55 2.07

Note. Using Dunn's procedure, the critical value for z, one-tailed, is 2.24.

*p < .05.

Premorbid Intellectual Abilities

499

However, even though the mean score estimates achieved the predicted pattern of results for each group, these patterns do not establish their validity. Because mean IQ scores are typically lower in brain-damaged groups, similar patterns may be obtained by arbitrarily assigning some number near 100 to predict IQ scores. Thus, any test of validity for an estimator also requires that it correlate with the score it is predicting for nonbrain-injured subjects. In the present study, the expectation that the estimates would correlate with WAIS-R IQ scores in the control group was not supported using the ICS, but was supported using the B84 and B86 equations. Thus, the current research failed to replicate the Schlottmann and Johnsen (1991 ) findings supporting the use of the ICS as an accurate estimator of intellectual functioning. It should be pointed out, however, that the correlations in the present study were based on only 20 pairs of observations. Given the rather small sample size, further research may be needed to confirm these results. It should also be noted that the ICS does not take any demographic variables into account. Comparison of the control groups in the current study and in the original Johnsen (1987) study suggests that the ICS may be more effective for males and/or urban subjects. It also may be sensitive to occupational choice and region of residence. Calculation of squared semipartial correlations demonstrated that the amount of variance accounted for on VIQ increased from 45% using only age, gender, and education to 63% when the ICS was added in. Although this increase was not statistically significant, it represents a considerable improvement over the ICS verbal results obtained when no demographic variables were taken into account. These results raise the question of whether a more robust estimator of VIQ could be created by formulating regression equations that take into account both different demographic factors and the type of individual characteristics tapped by the ICS. At the present time, however, it cannot be concluded that the ICS is a valid predictor of IQ. Any future research with the ICS should explore its reliability and validity across demographically different samples to ascertain whether there are certain populations in which its use is valid. Support is provided in the current research for the validity of the B84 and B86 as estimators of premorbid Verbal and Full Scale IQ. Both of these procedures correlated significantly with WAIS-R VIQ and FSIQ for controls. Other research (Schlottmann & Johnsen, 1991; Johnsen, 1987; Eppinger et al., 1987) has consistently found significant correlations for these equations with WAIS-R VIQ and FSIQ in nonbrain-injured subjects. Correlations between the B84 or B86 and WAIS-R PIQ have been more variable, ranging from a nonsignificant .30 in Johnsen's (1987) study to a highly significant .60 in the Eppinger et al. (1987) investigation. The repeated findings of significant correlations and, thus, of a relationship between WAIS-R VIQs and FSIQs with the B84 and B86 estimates suggest reliability and validity of these two regression equations over different samples. There appears to be somewhat more variability involved in using either the B84 or B86 to predict WAIS-R PIQ. Inspection of the B84 and B86 estimates suggests little difference between the degree of correlation that the two equations had with WAIS-R IQs. Thus, both equations predicting V1Q accounted for 27% of the variance. On FSIQ, the B84 accounted for 23% of the variance, whereas the B86 showed a very slight advantage, accounting for 27% of the variance. Similarly, in both classification analyses, the B86 again showed a very slight advantage over the B84, both with respect to correctly classifying subjects and with respect to giving a lower rate of false negatives and false positives. Results of the classification analysis to investigate the ability of the B84 and B86 to identify subjects with right, left, diffuse, or no brain injury indicated that both estimation procedures classified controls and diffuse brain-damaged subjects at a significantly better than chance level. However, neither procedure classified subjects with either right- or left-hemisphere lateralized injuries better than chance. The second classification analysis in which subjects were categorized as either brain damaged or nonbrain damaged revealed that the

500

S. A. Perez et al.

B84 and B86 correctly classified 80% and 83.75% of all cases, respectively. Neither set of equations classified brain-damaged subjects at a better than chance level, even though the majority of these subjects were classified correctly, perhaps a result of the rather small sample size used. On the other hand, the percentage of controls correctly classified by the B84 and B86, respectively, was 90% and 95%, both of which were significantly better than chance. Thus, both of the demographic regression equations were better able to classify controls as opposed to brain damaged subjects. It may seem that both sets of equations are more likely to produce false negatives than false positives. The pattern of false negatives and false positives, however, is affected by the decision rule that is used. In this study, a difference of 1 Standard Error of Estimate or better was used to identify brain damaged subjects. If a difference of .5 had been used, more false positives would probably have occurred. Judging from the results of these classification analyses and the correlational findings, caution is advised if using either of the demographic regression equations clinically, as there is much room for error in the individual case. If they are used clinically, it should be with an understanding of the potential for false positives and false negatives in individual cases. Of interest with relation to this topic is Eppinger et al.'s (1987) discussion regarding the use of difference scores with different cutoffs for individual cases. Previous studies (Eppinger et al., 1987; Johnsen, 1987; Schlottmann & Johnsen, 1991) have found that the B84 and/or B86 tend to overestimate WAIS-R IQ scores. In the current study, the B84 and B86 both produced good approximations of the scores being estimated. In the Eppinger et al. (1987) study, it was thought that the overestimation might be due to using controls who had been referred for neuropsychological assessment, but who had not received a neurological diagnosis. The current research used controls without any known psychiatric or neurological difficulties. This strategy appears to reduce the likelihood of overestimation by the B84 and B86 regression equations. By using acute and chronic brain-damaged groups, an attempt was made to investigate the effects of time since onset of injury upon the estimators. However, analyses indicated there were no significant effects for duration. This is likely to have been the result of an inability to control for the initial severity of the injury. In other words, the more severely damaged subjects were less likely to be testable in the acute stages. As expected, the left, right, and diffuse brain-damaged groups showed the expected pattern of results on WAIS-R Verbal and Performance IQ, reflecting the effects of lateralization of injury. However, it was also found that all of the brain-damaged groups obtained significantly lower scores than the control group on Verbal, Performance, and Full Scale IQ. Thus, it is concluded that left-hemisphere injury resulted in more severe, but not exclusive, impairment of verbal abilities, whereas right-hemisphere injury resulted in more severe, but not exclusive impairment of nonverbal abilities. In light of these findings, it would be simplistic to assume that the estimates would overpredict certain abilities, but not others, in cases of lateralized injury. Rather, what occurs is that the more severely impaired abilities are overestimated to a greater degree than are other less severely impaired abilities. In cases of lateralized left or right-damage, the estimators would tend to over predict both WAIS-R VIQ and PIQ, although one would be overpredicted to a greater extent than the other. This concept would be difficult to apply in a practical manner, such as that of making decisions about the lateralization of brain injury in individual cases. As can be observed in the first classification analysis conducted in the present study, the estimation procedures were not able to accurately predict group membership for either the right or left hemisphere-damaged groups at a better than chance level. Thus, it is concluded that lateralized effects are more easily and accurately observed by considering the differences between WAIS-R VIQ and PIQ together with other test data and patient history. It is recommended that none of the estimation procedures be used to make decisions about the lateralization of brain injury, particularly in individual cases.

Premorbid Intellectual Abilities

501

REFERENCES Barona, A., & Chastain, R. L. (1986). An improved estimate of premorbid IQ for blacks and whites on the WAISR. The International Journal of Clinical Neuropsychology, 8(4), 169-173. Barona, A., Reynolds, C. R., & Chastain, R. (1984). A demographically based index of premorbid intelligence for the WA1S-R. Journal of Consulting and Clinical Psychology, 52(5), 885-887. Chelune, G. J., Ferguson, W., & Moehle, K. (1986). The role of standard cognitive and personality tests in neuropsychological assessment. In T. Incagnoli, G. C~ldstein, & C. J. Golden (Eds.), Clinical application ofneuropsychological test batteries (pp. 75-119). New York: Plenum Press. Eppinger, M. G., Craig, R L., Adams, R. L., & Parsons, O. A. (1987). The WAIS-R index for estimating premorbid intelligence: Cross-validation and clinical utility. Journal of Consulting and Clinical Psychology, 55(I ), 86-90. Gentry, B. (I 982). A feasibility study for the development of an intellectual correlate scale. Unpublished thesis, Oklahoma State University, Oklahoma. Hubert),, C. J. (1984). Issues in the use and interpretation of discriminant analysis. Psychological Bulletin, 95(I). 156-171. Hunt, W. L. (1949). The relative rates of decline of Wechsler-Bellevue "hold" and "don't hold" tests. Journal of Consulting Psychology, 13, 440-443. Johnsen, D. E. (1987). The development and use of an intellectual correlates scale in the prediction of premorbid intelligence in adults. Unpublished doctoral dissertation, Oklahoma State University, Oklahoma. Kaufman, A. S. (1990). Assessing adolescent and adult intelligence. Boston: Allyn Bacon. Klesges, R. C., & Troster, A. 1. (1987). A review of premorbid indices of intellectual and neuropsychological functioning: What have we learned in the past five years. International Journal of Clinical Neuropsychology, 9( 1), 1-11. Lezak, M. D. (1983). Neuropsychological assessment (2nd ed.). New York: Oxford University Press. Milligan, G. W., Wong, D. S., & Thompson, P. A. (1987). Robustness properties of nonorthogonal analysis of variance. Psychological Bulletin, 101,464-470. Schlottmann, R. S., & Johnsen, D. E. (1991). The intellectual correlates scale and the prediction of premorbid intelligence in brain-damaged adulm. Archives of Clinical Neuropsychology, 6, 363-374.