Demographically Corrected Norms for African Americans and Caucasians on the Hopkins Verbal Learning Test-Revised, Brief Visuospatial Memory Test-Revised, Stroop Color and Word Test, and Wisconsin Card Sorting Test 64-Card Version

Abstract

Memory and executive functioning are two important components of clinical neuropsychological (NP) practice and research. Multiple demographic factors are known to affect performance differentially on most NP tests, but adequate normative corrections, inclusive of race/ethnicity, are not available for many widely used instruments. This study compared demographic contributions for widely used tests of verbal and visual learning and memory (Brief Visual Memory Test-Revised, Hopkins Verbal Memory Test-Revised), and executive functioning (Stroop Color and Word Test, Wisconsin Card Sorting Test-64) in groups of healthy Caucasians (n = 143) and African-Americans (n = 103). Demographic factors of age, education, gender, and race/ethnicity were found to be significant factors on some indices of all four tests. The magnitude of demographic contributions (especially age) was greater for African-Americans than Caucasians on most measures. New, demographically corrected T-score formulas were calculated for each race/ethnicity. The rates of NP impairment using previously published normative standards significantly overestimated NP impairment in African-Americans. Utilizing the new demographic corrections developed and presented herein, NP impairment rates were comparable between the two race/ethnicities and unrelated to the other demographic characteristics (age, education, gender) in either race/ethnicity group. Findings support the need to consider extended demographic contributions to neuropsychological test performance in clinical and research settings.

Introduction

Learning, memory and executive functioning are core components of comprehensive neuropsychological (NP) assessment batteries. Accurate classification of NP impairment in these domains is especially important for the differential diagnosis of many neurologic conditions. Unfortunately, some of the most widely used neuropsychological tests do not have available norms that are corrected for race/ethnicity differences, despite research showing that differential ethnicity backgrounds affect NP performance, along with other demographic variables such as age, education, and gender (Heaton, Miller, Taylor, & Grant, 2004). Inadequate normative sampling and standards may lead to neuropsychological misclassification and may particularly contribute to misdiagnosis of African-Americans.

Researchers have begun examining demographic influences on learning and memory performance in an effort to produce normative standards among minority groups such as in African-Americans (e.g., California Verbal Learning Test -CVLT)[Norman et al., 2000]; and the Third Edition of the Wechsler Memory Scale, WMS-III [Heaton, Taylor, & Manly, 2003],and in Spanish-speaking Hispanics (Hopkins Verbal Learning Test-Revised—HVLT-Rand Brief Visuospatial Memory Test-Revised—BVMT-R) (Cherner, et al., 2007). However, data on a wider range of neuropsychological tasks are lacking. Regardless of the domain or racial group under study, race/ethnicity typically influences scores on NP measures (Manly, Schupf, Tang, & Stern, 2005).

Several studies have demonstrated lower NP performance among African-Americans as compared to Caucasians on a wide variety of NP measures (Diehr, et al., 2003; Diehr, Heaton, Miller, & Grant, 1998; Gladsjo, et al., 1999; Heaton, et al., 2004; Heaton, et al., 2003; Norman, 2000; Rilling, et al., 2005). Importantly, it has been shown that these differences persists even when groups were matched for other demographic factors, including age, gender, education, and reading ability to a lesser extent (Manly, et al., 2005).

Accurate classification of the level of NP impairment in diverse racial groups has pragmatic clinical relevance to neuropsychologists. Without race/ethnicity-corrected scores in the clinical setting, a substantial number of normal African-Americans patients might be incorrectly classified as neuropsychologically impaired, and misdiagnosed. For example, Norman et al. (2000) demonstrated that 46%of African-Americans were classified as NP impaired (i.e., NP test T-score< 40) on the California Verbal Learning Test (CVLT) Trials 1–5 using the original Delis et al.(Delis, Kramer, Kaplan, & Ober, 1987) norms, which were based upon a predominantly Caucasian standardization sample. When CVLT norms were corrected for age, gender, and education, but not race,36% of African-Americans were still classified as NP impaired. Once race was sufficiently accounted for in the equation, only 17.8% of African Americans were classified as neuropsychologically impaired demonstrating that the new CVLT norms clearly improved the proportion of individuals scoring greater than one standard deviation below the mean. Misdiagnoses for neurodegenerative disorders or other conditions that affect brain functions have serious implications in terms of public health consequences as well as social and healthcare consequences for the patients and their families. Accurate classification of NP-impairment among African Americans is equally important in research settings for similar reasons.

Most existing normative data that are published in test manuals, lack information about race/ethnicity influences on test performance. The Hopkins Verbal Learning Test-Revised (HVLT-R) is a widely used task of verbal learning and memory (Brandt & Benedict, 2001), using 12 words belonging to three semantic categories. Six alternate forms facilitate reducing practice effects on repeated administrations. The standardization group’s age range was from 15 to 92 (M = 59.0, SD = 18.6) and education ranged from 2 to 20 years (M = 13.4, SD = 2.9); 79% were women. The normative sample for the HVLT-R included 1,179 adults; however, racial/ethnicity demographics were not provided. In the HVLT-R manual, stepwise multiple regression examined the effects of age, education, and gender for HVLT-R Total Recall, Delayed Recall, Percent Retained, and Recognition Discrimination. Age accounted for a considerable amount of variance, but education and gender were not found to significantly contribute to test performance. Cherner et al. (Cherner, et al., 2007) contend that a limitation of the original norms was that the reference group was highly educated and had suboptimal representation of low levels of education. Because of this reference group limitation, the rate of NP impairment may be erroneously elevated among lower educated persons.

The Brief Visual Memory Test – Revised (BVMT-R) is a short task of visual memory (Benedict, 1997). As with the HVLT-R, there are six different versions that allow for repeat testing with reduced practice effects. Similar to the HVLT-R, the manual describes a standardization group of 588 healthy English-speaking adults (171 college students and 471 community-dwelling participants) between the ages of 18 and 79 (M = 38.6, SD = 18.0) and with a mean education of 13.4 years (SD = 1.8). African-Americans accounted for 14.5% of the standardization sample; however, the authors did not provide information concerning whether and how race/ethnicity related to BVMT-R performance. The BVMT-R and HVLT-R produce indices of Total Recall, Delayed Recall, Percent Retained, and a Recognition Discrimination Index.

The Stroop Color and Word Test consists of speeded trials of Word Reading, Color Naming, and Color-Word Interference. Numerous versions of the Stroop exist, and the version used in the current study assigns a score for each trial based on the number of words read or colors named in forty-five seconds (Golden, 1978). The normative sample mentioned in the 2002 manual (Golden & Freshwater, 2002) includes the previous normative group (n = 100) from the original manual (Golden, 1978) as well as 300 additional cases collected between 1977 and 1997 (Golden & Freshwater, 2002). Age and education showed significant associations with Stroop scores and as such, the manual includes predicted scores for each trial based on these two demographic characteristics. Gender effects on the Stroop have been examined, but have been found to be inconsistent and confounded by sampling concerns; however, the racial characteristics of the original or total normative samples were not described.

The Wisconsin Card Sorting Test – 64 Card Version (WCST-64) is a computerized test of executive function that requires strategic planning and the ability to use environmental feedback to shift cognitive set (Kongs, Thompson, Iverson, & Heaton, 1993). The normative sample for the WCST-64 consisted of 445 adults ages 18–89 (M = 49.83 SD = 17.92). Education ranged from 6 to 20 years (M = 14.95, SD = 2.97) and 23% of the sample was female. Unfortunately, information about race/ethnicity was not routinely collected and therefore was not available for analysis. The manual states that hierarchical polynomial regressions were used to examine the effects of age, gender, and education. Age demonstrated a significant quadratic relationship with WSCT-64 scores and accounted for 1.4% to 18.9% of the variance in scores. Education accounted 1.3% to 7.7% of the variance in scores after adjusting for age. There were no significant gender effects after accounting for age and education.

This current study was designed to provide improved, demographically corrected normative standards among healthy samples of African-Americans and Caucasians on the HVLT-R, BVMT-R, Stroop Color and Word Test, and Wisconsin Card Sorting Test – 64 Card Version. The project has two specific aims: 1) To analyze the effects of demographic variables, including race/ethnicity (i.e., African-American and Caucasian) on test performance and classification accuracy (normal vs. abnormal), and 2) To develop normative equations that correct for all relevant demographic characteristics (age, education, gender, and race/ethnicity) to provide a more accurate classification of NP performance. We predict that Caucasian vs. African-American race/ethnicity will significantly contribute to NP performance, and that these differences will support the assertion that verbal and visual learning and memory as well as executive function measures require race/ethnicity corrections in order to correctly categorize NP impairment among African Americans (Manly & Echemendia, 2007).

Methods

Participants

The sample consisted of 246 healthy individuals recruited as comparison participants (HIV uninfected controls) in a longitudinal study of HIV infected participants at the University of California, San Diego (UCSD)HIV Neurobehavioral Research Center (HNRC). One hundred forty three participants self identified as Caucasian and 103 self identified as African-American. Trained research associates used structured interviews and administered screening questionnaires to potential participants to assess inclusion/exclusion criteria prior to study enrollment. Exclusionary criteria for all subjects included any history of neurological disorders, current substance use disorders, and other conditions (e.g., psychiatric disorder with psychotic features, medications with CNS effects) known to affect neurocognitive performance. The UCSD Human Research Protections Program approved the protocol. Demographic information is presented in Table 1. The cross-sectional, stratified sample ranged in age from 20 to 65. The two samples (African American and Caucasian) did not differ significantly in terms of age or education, but the Caucasian group contained a smaller proportion of females (31% vs. 50%).

Table 1A.

Demographic information in the two groups (Mean, SD)

Mean (SD) [Range]	Caucasian (n=143)	African American (n=103)	p
Age	37.6 (12.3) [20–66]	40.6 (12.3) [20–69]	0.06
Education	14.1 (2.4) [8–20]	13.8(2.1) [8–19]	0.37
Sex (% Female)	31%	50%	0.003

Participants were asked to self identify their own race/ethnicity and this identification was used to define the African American and Caucasian groups used in this study. Years of education were determined using a previously defined and standardized procedure where education level ranges from 0–20 based on number of years of schooling completed (Heaton, et al., 2004). For example, a high school graduate receives 12 years of education and a person with a bachelor’s degree receives 16 years of education.

Neuropsychological Assessment

Participants completed an NP test battery of which a subset of two memory and two executive function tests were examined for this study, because these tests lacked race/ethnicity corrections as compared to other tests in the battery. Trained psychometrists following instructions from the respective manuals completed administration and scoring. Analyzed measures included Form A of the Hopkins Verbal Learning Test-Revised (Brandt & Benedict, 2001), Display A from the Brief Visual Memory Test-Revised (Benedict, 1997), the Stroop Color-Word Interference Test (Golden, 1978), and the Wisconsin Card Sorting Test-64 Computer Version (Heaton, Chelune, Talley, Kay, & Curtiss, 1993). We evaluated Total Recall across three learning trials and Delayed Recall for the HVLT-R and BVMT-R. Additionally, total numbers of correct items identified with the 45-second trials were analyzed for StroopWord Reading, Color Naming, and Color-Word formats. For the WCST, scores analyzed included Total Errors, Perseverative Errors, and Conceptual Level Responses.

Data Analysis

The distributions of all scores were examined. Although distributions of test raw scores were non-normal, parametric statistics were confirmed with non-parametric versions of the same statistical comparisons, and tails of distributions were similar between racial groups assuming symmetry in impairment rates. Effect sizes were measured with the unbiased Cohen’s d (Hedges & Olkin, 1985). This study was powered to detect a small effect size.

In the first step, African-American and Caucasian group scores were compared to analyze the effects of race/ethnicity on test performance. Next, linear regression was used to examine the effects of age, education, and gender; this was done separately for African American and Caucasian groups because it was determined that they had somewhat different age effects and (to a lesser degree) education effects. Partial regressions were then run to examine the independent contribution of age, education, and gender on measures in each group (Caucasian and African-American).

HVLT-R, BVMT-R, Stroop, and WCST-64 raw scores for the total subject group were converted into quantiles and mapped into the corresponding quantiles of a standard normal distribution. These scores were then converted into normalized scaled scores with a mean of 10 and standard deviation of 3. We used a subset of individuals (n=208) from the present study and some additional normal subjects from other ethnicities to create census-matched subset of individuals to generate the scaled scores as described below, but results from other ethnicities were not used in subsequent analyses that focused on African Americans and Caucasians. The rationale for adding these additional individuals for raw score to scaled score conversions was to reflect in the scaled scores the major ethnic group composition reported in the 2000 US census. These individuals met the same screening procedures as the study population. The resulting census matched proportions of race/ethnicity categories were 68.7% Caucasian, 13.5% African-American, 13.0% Hispanic, and 4.8% other race/ethnicities. Scaled score conversion tables for all variables are presented in Tables 2–4.

Table 2.

Raw-to-Scaled Score conversions for the BVMT-R and HVLT-R

	BVMT-R		HVLT-R
Scaled	Total Recall Raw	Delayed Recall Raw	Total Recall Raw	Delayed Recall Raw
17	36		36
16
15	34–35		35
14	33	12	34
13	32		33	12
12	30–31		31–32
11	28–29	11	30	11
10	26–27		29
9	24–25	10	27–28	10
8	21–23	9	26
7	19–20	8	24–25	9
6	16–18	7	22–23	8
5	14–15	5–6	21	7
4	10–13	4	20	5–6
3	0–9	3	16–19
2		0–2	0–15	4
1				0–3

Table 4.

Raw-to-Scaled Score conversions for the WCST-64

Scaled	Total Errors	Perseverative Errors	Conceptual Level Responses
18		0–3
17	0–6		≥58
16	7		57
15	8		56
14		4	54–55
13	9–10		53
12	11	5	51–52
11	12	6	49–50
10	13–15	7	45–48
9	16–19	8	39–44
8	20–22	9–10	34–38
7	23–28	11–13	28–33
6	29–32	14–15	20–27
5	33–35	16–18	16–19
4	36–39	19–26	13–15
3	40–48	26–41	6–12
2	≥49	≥42	≤5
1

In the next step, fractional polynomial multiple regression was employed to develop demographically-corrected prediction equations on the Caucasian and African American samples (respective n’s = 143 and 103) for each NP test scaled score using the methods outlined by Royston and Altman (Royston & Altman, 1994; also see Heaton, et al.,2004, and Cherner et al., 2007). Separate regressions were run for each race/ethnicity, and the predictors included age, education, and sex. The fractional polynomial method developed by Royston and Altman (1994) uses an interactive algorithm to evaluate the influence of combinations of predictors with predetermined exponents (−2, −1, −0.5, 0, 0.5, 1, 2, 3) (the coefficient of 0 stands for the natural logarithm transformation). The algorithm compares all sets of predictors using these transformations to generate the final optimal fit. The residuals from the optimal regression equations were converted to T-scores with a mean of 50 and a standard deviation of 10. As designed, the resultant T-scores are not correlated with age, sex, or education for either racial group.

Results

In the first step, African American and Caucasian raw scores on each of the neuropsychological measures examined in this study were compared to analyze the effects of race/ethnicity. Table 5 demonstrates significant Caucasian and African American differences on all measures, such that Caucasians performed better in each instance. Table 5 also depicts medium to large effect sizes on most learning, memory and executive functioning indices; the only exceptions were small to medium effect sizes on HVLT-Delayed Recall and Stroop Color Naming and Word Reading.

Table 5.

Neuropsychological test performance (raw scores) in Caucasians and African-Americans (Mean, SD)

	Caucasian (n=143)	African-American (n=103)	p	Cohen’s d
BVMT-R Total Recall	26.5 (5.9)	22.7 (6.8)	.0003	−0.60
BVMT-R Delayed Recall	10.2 (1.7)	8.7 (2.4)	<.0001	−0.74
HVLT-R Total Recall	29.2 (3.9)	26.8 (4.9)	.0002	−0.55
HVLT-R Delayed Recall	10.4 (1.9)	9.4 (2.3)	.0016	−0.48
Stroop Word Reading	101.9 (14.4)	96.2 (16.9)	.007	−0.37
Stroop Color Naming	76.4 (10.8)	70.8 (13.0)	.0008	−0.47
StroopColor-Word	45.0 (9.5)	38.2 (10.2)	<.0001	−0.69
WCST-64 Total Errors	15.6 (7.8)	22.1 (10.2)	<.0001	0.73
WCST-64 Perseverative Errors	7.6 (3.9)	11.0 (4.2)	.0002	0.63
WCST-64Conceptual Level Responses	44.3 (11.3)	35.5 (14.3)	<.0001	−0.69

For each of the 10 test scores, stepwise linear regressions were then conducted separately for each group (African American & Caucasian) to determine the proportion of variance accounted for by age, education, and gender (Tables 6 & 7). None of the fractional polynomials were significant predictors.

Table 6.

Full model R² and partial R² for the effect of age, education and gender in each group on memory measures.

	Caucasian (n=143)			African American (n=103)
	R2	Partial R2	95% CI	R2	PartialR2	95% CI
BVMT-R Total Recall	.11**		.02, .20	.28***		.14, .42
Age		.05**	.00, .12		.21***	.07, .35
Education		.05**	.00, .12		.05**	.00, .13
Sex		.02	.00, .06		.03*	.00, .09
BVMT-R Delayed Recall	.06*		.00, .13	.29***		.15, .43
Age		.03*	.00, .08		.24***	.10, .38
Education		.04*	.00, .10		.03*	.00, .09
Sex		.00	.00, .02		.03*	.00, .09
HVLT-R Total Recall	.19***		.07, .30	.15***		.03, .27
Age		.00	.00, .02		.07**	.00, .16
Education		.18***	.06, .29		.05*	.00, .13
Sex		.03*	.00, .08		.03*	.00, .09
HVLT-R Delayed Recall	.18***		.06, .29	.14***		.02, .26
Age		.01	.00, .04		.05*	.00, .13
Education		.14***	.04, .24		.08**	.00, .18
Sex		.05**	.00, .12		.01	.00, .05

^*p≤ .05

^**p≤ .01

^***p≤ .001

Table 7.

Full model R² and partial R² for the effect of age, education and gender in each group on executive measures.

	Caucasian (n=143)			African-American (n=103)
	R2	Partial R2	95% CI	R2	Partial R2	95% CI
Stroop Measures
Word Reading	.01		.00, .04	.20***		07, .33
Age		.00	.00, .02		.12***	.00, .24
Education		.00	.00, .02		.06*	.00, .15
Sex		.00	.00, .02		.03	.00, .09
Color Naming	.01		.00, .04	.26***		.12, 40
Age		.00	.00, .02		.16***	.03,.29
Education		.01	.00, .04		.04*	.00, .11
Sex		.00	.00, .02		.05**	.00, .13
Color-Word	.15***		.04, .25	.39***		.25, .53
Age		.11***	.01, .20		.28***	.14 .42
Education		.04*	.00, .10		.03*	.00, .09
Sex		.00	.00, .02		.10***	.00, .21
WCST-64 Measures
Total Errors	.22***		.10, .34	.19**		.06, .32
Age		.17***	.06, .28		.13**	.01, .25
Education		.07***	.00, .15		.06**	.00, .15
Sex		.00	.00, .02		.00	.00, .03
Perseverative Errors	.20***		.09, .31	.14**		.02, .26
Age		.17***	.06, .28		.09**	.00, .19
Education		.05**	.00, .12		.05*	.00, .13
Sex		.00	.00, .02		.00	.00, .03
Conceptual Level Responses	.20***		.09, .31	.20***		.07, .33
Age		.15***	.04, .26		.13***	.01, .25
Education		.07***	.00, .15		.07**	.00, .16
Sex		.00	.00, .02		.00	.00, .03

^*p≤.05

^**p≤ .01

^***p≤ .001

Memory

Table 6 shows information related to the demographic influences on learning and memory performance in Caucasians and African Americans independently. When considering the partial R² results, only the African-American group showed a significant effect of age, and this was true for all measures (especially robust for BVMT-R measures). Total demographic effects (R²s) were higher for African Americans due to greater age effects on the BVMT-R, whereas more comparable effects were seen for the HVLT-R. Although education was a significant independent predictor of memory test performance for all measures in both groups, the education effects on the verbal (HVLT-R) measures were especially robust for the Caucasian group. Gender effects were absent or modest for both groups on most measures, with women performing better, and there were no systematic differences for the Caucasians versus African Americans.

Executive Functioning

Comparable results for Stroop and WCST-64 measures are presented in Table 7. As was the case for measures of visual learning and memory (BVMT-R), only the African American groups showed very large independent effects of age on all of the Stroop indices (Word Reading, Color Naming, and Color-Word). Only the African American group also showed significant gender effects on Stroop Color-Word (Interference condition) and Color Naming, with women performing faster. On the WCST-64 measures, both race/ethnicity groups demonstrated medium sized age effects (typically somewhat larger for Caucasians), and usually small to medium education effects. Neither racial/ethnicity group showed gender effect on this test.

Normative T-Score Derivation

As described in the Methods section, fractional polynomial regression analyses were conducted to derive normative scores that would correct for the observed demographic effects on normal test performance. This procedure began with the conversion of raw scores to normalized scaled scores (mean = 10, SD = 3) on all test measures (see Tables 2–4 for these conversions).

To examine the diagnostic (“normal” versus “abnormal”) classification accuracy of the new T-score conversions with more complete demographic corrections, we compared the impairment rates in both samples with those using previously published normative data (Benedict, 1997; Brandt & Benedict, 2001; Golden & Freshwater, 2002; Kongs, et al., 1993) that did not correct for race/ethnicity. The formulas used to generate the results for the new T-scores are included in Appendix A. Subjects were considered impaired if their T-Score was less than 40(Heaton, et al., 2004; Taylor & Heaton, 2001).

Figure 1 shows the results for the African American group. When applying previously published normative corrections to this sample, 24–49% of normal individuals were classified as NP impaired depending on the test score examined. Using our newly generated normative data the impairment rates significantly improved and ranged from 13–16%. The impairment rates for the African American sample with the previously published norms are significantly greater than what would be expected from the normal distribution with the selected 1 SD cutoff (Golden, 1978). All comparisons of impairment rates among African Americans using previously published normative corrections as compared to the newly generated normative corrections were statistically significant with the exception of the HVLT-R Delayed Recall measure that approached significance (p=0.08).

Figure 1.

Percent of normal African American sample classified as “impaired” (1 SD cutoff) by published norms versus new, demographically corrected norms.

*p≤.05

**p≤ .01

***p≤ .001

The new normative correction formulas improved the consistency of impairment rates across test scores for the Caucasian sample as well (see Figure 2). Impairment rates for these norms ranged from 12 to 17% as compared to 8 to 26% using previously published normative data. The newly developed WCST-64 norms produced impairment rates more aligned with the expected impairment rates and were significantly lower than impairment rates with previously published norms on all WCST-64 indices.

Figure 2.

Percent of normal Caucasian sample classified as “impaired” (1 SD cutoff) by published norms versus new, demographically corrected norms.

*p≤ .05

**p≤ .01

***p≤ .001

Discussion

This study complements previous literature on demographic corrections for neuropsychological test norms by examining a broader range of memory and executive functioning measures and specifically examining the effect of African American versus Caucasian race/ethnicity on test performance. These findings strongly support the use of separate norms for African-American and Caucasian examinees on the tests used here and, when combined with previously published results in the same ability domains, on learning, memory and executive functioning measures more generally. Consistent with prior findings on the Wechsler Intelligence and Memory Scales (Heaton, et al., 2003) and expanded Halstead-Reitan Battery (Heaton, et al., 2004), we found, in our sample of 103 African-Americans and 143 Caucasians, that African-American participants obtained lower raw scores on visual and verbal learning and memory and executive functioning measures.

There are multiple background differences between African American and Caucasian adults within U.S. society today that may place African Americans at a disadvantage on standardized NP testing. The observed raw NP score differences may be consistent with disparities in quality of formal and informal educational experiences; however, other factors may also contribute to these discrepancies. It is considered unlikely that race has a direct causal effect on differences in adult cognition, so race/ethnicity is viewed as a proxy for other factors, much like has been discussed about education (Manly, Byrd, Touradji, & Stern, 2004). Factors potentially contributing to raw NP score differences between African American and Caucasian groups may include academic exposure, education quality, academic resources, acculturation, socioeconomic status, social exposure, “test wiseness”, societal discrimination (Byrd, Sanchez, & Manly, 2005; Manly, et al., 2004) and lifelong experiences contributing to low group and self-expectations (Steele & Aronson, 1995).

There are few opportunities in the literature to compare our raw score results with those reported by other investigators. Whereas this study found about a 2-point (raw score) difference on HVLT-R Total Recall performances between Caucasians and African-Americans, Morgan et al. (Morgan, Marsiske, & Whitfield, 2008) found a 4-point difference and less variability. The current study demonstrated moderate to large race/ethnicity effect sizes, but the raw score differences between Caucasians and African Americans do not seem to be especially large (e.g., an average of only 1.5 points on BVMT-R Delayed Recall is associated with a medium to large effect size; see Table 5). However, these differences were sufficiently robust to cause unacceptably large “impairment” classification rates in the African American sample (Figure 1).

Although concerns might be raised that the method of raced-based norming could “overcorrect” performances of neurologically impaired African Americans (making them less sensitive to disease), this could be said as well for norms that correct for older age, lower education levels, or any demographic characteristic that is associated with lower test performance in normal people. In our view, the most important function of norms is to maintain an acceptable and consistent level of diagnostic specificity (accuracy in classifying normal people as normal) for people regardless of their demographic characteristics. Our data suggest that the norms presented here result in rates of impairment that are comparable, and are within statistical expectations for a healthy population, for both our Caucasian and African American participants.

As addressed by Byrd et al. (Byrd, et al., 2005) and others, the term “race” is an arbitrary distinction and difficult to operationalize. Often race is based on skin color and self-identification. As Gasquoine (2009) notes, race is a social definition rather than a scientific classification and race is not homogeneous. Devising ways to understand the influences of ethnicity and race on NP tests will become increasingly complex as rates of self-identified multiracial individuals rise.

Given the unclear relationship of “race” on cognition, some suggest recording, quantifying, and modeling the effects of all background factors that can influence cognitive development and test performance. Gasquoine (Gasquoine, 2009) and others have advocated that an alternative approach to race/ethnicity-based norms is to estimate preexisting neuropsychological status based on a case-by-case basis from regular normative tables. On the other hand, Gasquoine acknowledges that there is little empirical support for this technique, and there is no agreed upon method for establishing NP status on a case-by-case basis. Furthermore, accurate retrospective collection of such complex data across the lifespan is very difficult (Byrd, et al., 2005).

Also, a subjective interpretation of cognitive deficit will most likely have the effect of wide variations in the impairment classifications of minorities between different clinical neuropsychologists. Instead, the use of the more general race/ethnicity proxy (with all its shortcomings) in normative corrections should at least enhance consistency/reliability and may greatly reduce the probability of incorrectly attributing cognitive and possibly central nervous system abnormalities to normal African Americans.

Of course, clinical interpretation of neuropsychological data should not strictly rely upon use of norms, but also consider the appropriateness of available norms in relation to each person’s background, including social, educational and medial history, and other factors (i.e., psychiatric, substance use, etc.). In particular, diagnostic sensitivity and specificity are likely to vary when norms are applied to people whose backgrounds differ significantly from those represented in the normative sample populations (Heaton et al., 2004).

In addition, it is important to note that demographically corrected norms are intended to reflect the difference between current performance and a best estimate of the person’s expected “normal” performance (i.e., in the absence of CNS abnormality). Such norms are less appropriate, at best, when the goal is to determine the person’s absolute level of ability (e.g., in relation to requirements of specific everyday tasks and activities).

Following the derivation of separate normative equations and confirming adequate normative distributions, we found that our new demographically corrected formulas provided significantly improved impairment estimates. These data suggest that scores that have not been corrected for race/ethnicity classify 31–32% of the African-American sample with visual learning and memory impairment, 25–26% as having verbal learning and memory impairment, and 24–49% as having executive dysfunction. These percentages are substantially higher than expected values in any normative population. In contrast, when the African-American scores were corrected for race/ethnicity, the average impairment frequencies dropped to expected levels. The over estimation of impairment with existing normative data can lead to misclassification and/or misdiagnosis of African American individuals and can have serious negative consequences for the patients and their families. Misdiagnosis and misclassification is problematic in clinical, forensic, and research applications of neuropsychology; however, few NP norms account for these demographic variables.

The present study also demonstrated that the demographic contributions of age, education and gender to NP test performances were somewhat different for African Americans as compared to Caucasians. The contribution of age tended to be stronger for African-American participants on the BVMT-R, HVLT-R, and Stroop tests, but less pronounced on the WCST-64. The current study was not designed to explore why demographic factors exhibit stronger influences among African Americans, although large age effects for African Americans as compared to Caucasians have been observed in other large U.S. samples and on other neuropsychological tests (Heaton et al., 2004). Because these differential effects of demographics are not well understood, they require additional careful investigation (especially taking into account age related medical conditions and associated treatments that could differ across ethnicity groups).

The current study is limited, as are others of the same type, in terms of failing to provide insights into the factors that contribute to racial differences on these memory and executive function measures. As discussed earlier, Manly (1998) and others suggest that educational quality, exposure, and other factors might play a role in the poorer observed performance of African Americans on these neuropsychological tests. The amount of education may be less important than the quality of one’s education, as measured by reading scores. Dotson et al. (Dotson, Kitner-Triolo, Evans, & Zonderman, 2008) and Manly et al. (Manly, et al., 1999; Manly, Jacobs, Touradji, Small, & Stern, 2002) found that literacy was a better predication of cognitive scores than education. In an African-American sample, Dotson and colleagues (2008) used memory, naming, fluency, visuospatial, attention, and psychomotor scores and regressed them on sex, age, literacy, and education scores. They did not find a unique contribution of education after literacy was added to the model; however, this study only included African-American participants. The present study did not measure literacy, but as with previous studies, education was found to be a significant predictor to cognitive scores. The measurement of these factors remains elusive, however, as effects of educational opportunities and importance within the cultural experience, and other potentially important factors are complex and difficult to determine retrospectively (e.g., asking an adult about parental influences and early school experiences; Byrd et al., 2005). On the other hand, current attempts at understanding these factors are starting to emerge and multifactorial models involving psychological factors, stress factors, social and cognitive factors have been proposed (Mays et al., 2007). An additional complexity is that it is likely that some or all of the factors influencing NP test performance have changed over generations and continue to do so. For example, it is likely that the educational quality for 30 year-old and 60 year-old African Americans has been quite different (probably more so for than for Caucasians in the U.S.).

An additional limitation in this research is the ambiguity in classifying race or ethnicity. While “race” and “ethnicity” are often interchangeably used in this area of research, they are not equivalent terms. Given that there are no biological race/ethnicity markers, group identification has been pragmatically based on self-identification – and this is the approach that was used on the current study. Race is more than just skin color and there may be multiple ethnic groups within a race. Some argue that the inability to specifically identify and characterize race/ethnicity should preclude demographic corrections; however, even with this limitation, the current data demonstrate excessive rates of diagnostic error if clinicians use norms that are not corrected for race/ethnicity. In particular, our findings with the new T-score conversions suggest greater and more equal specificity, with regard to race/ethnicity, within the healthy population than was achieved by the published norms. Although we have no data concerning sensitivity of the new norms to CNS compromise, sensitivity also is likely to be more equivalent among demographic groups (e.g., (Heaton, Ryan, & Grant, 2009). Despite limitations, we believe that the current quantitative standards provide a substantial improvement for the classification of neurocognitive impairment status in self identified African-Americans.

Finally, it is important to acknowledge that our current sample size was relatively small, and we were unable to cross-validate the normative distribution with an independent sample. We recommend caution when using these normative data with individuals over age 60 or with other groups not well represented in our normative sample. There were relatively few individuals with less than 10 years of education enrolled in this study and therefore caution should be used when applying these normative corrections to persons with such low levels of education. In addition, all participants in this study were from the San Diego area and participants were carefully screened to exclude anyone with neuromedical or developmental histories suggesting any increased risk for CNS compromise. As such, generalizability of these results and associated normative standards to other, ostensibly normal, African American and Caucasian groups cannot be assumed. To partially address this question, we applied the demographically corrected norms in the WAIS-III/WMS-III/WIAT-II Scoring Assistant program (The Psychological Corporation, 1999; Heaton Taylor and Manly, 2003) to the current samples’ results on three WAIS-III subtests (Letter-Number Sequencing, Digit Symbol Coding, and Symbol Search). These latter norms were based upon a large, national standardization sample from all U.S. regions, and correct for all demographic variables that were examined in the current study (age, education, gender and African American versus Caucasian race/ethnicity). We reasoned that application of these norms to the current samples’ WAIS-III results would provide some indication of their representativeness of the much larger national sample. Ideally, the mean (SD) T-scores would approach 50 (10) and would not differ for the two race/ethnicity groups in the study.

For Letter-Number Sequencing, the mean (SD) T-scores were 53.0 (9.8) for our African American Group and 51.8 (9.4) for our Caucasian group (p-value for group difference = 0.35). On the WAIS-III Processing Speed Index (which combines Digit Symbol and Symbol Search), the respective scores were 54.6 (10.7) for our African American group and 52.6 (9.9) for our Caucasian group (p=.14). The fact that both of our race/ethnicity groups performed slightly better than the national standardization samples on these WAIS-III tests may reflect our (arguably) more stringent neuromedical screening procedures and/or slight regional differences. Also, however, these results indicate that, relative to normal expectations for African American and Caucasians in the U.S., our race/ethnicity groups performed comparably. This suggests that our groups’ findings on the memory and executive function tests are unlikely to be overestimating the race/ethnicity bias in the previously published norms.

Our results for the HVLT-R and BVMT-R are limited to Form A of these measures, and future studies will focus on assessing the need for specific corrections for all the multiple forms of these measures. Additionally, it is important to assess whether or not the demographic corrections can be validated in a clinical sample, showing equivalent results across the various demographic categories.

Table 1B.

Caucasian

Education	Age Range
	<30	30–39	40–49	50–59	60+
< 10	0	1	1	0	0
10–11	4	2	2	1	0
12	5	6	5	5	2
13–15	12	4	13	7	0
16	7	5	4	5	0
>16	2	6	6	6	0

Table 1C.

African American

Education	Age Range
	<30	30–39	40–49	50–59	60+
< 10	0	1	0	0	0
10–11	1	1	4	1	1
12	6	4	6	3	1
13–15	8	9	10	8	4
16	4	4	5	1	1
>16	0	3	1	2	2

Table 3.

Raw-to-Scaled Score conversions for the Stroop Color and Word Test

	Stroop Color and Word Test
Scaled	Word Reading Raw	Color Naming Raw	Color-Word Raw
18	≥145	≥107
17	134–144	100–106	≥65
16	128–133	97–99	63–64
15	123–127	93–96	59–62
14	118–122	89–92	56–58
13	114–117	85–88	53–55
12	109–113	80–84	49–52
11	106–108	76–79	46–48
10	101–105	74–75	42–45
9	97–100	70–73	39–41
8	89–96	66–69	36–38
7	83–88	62–65	32–35
6	77–82	58–61	29–31
5	71–76	49–57	25–28
4	67–70	43–48	22–24
3	66	40–42	0–21
2	<66	0–39

Appendix A: Normative Formulas for Caucasians and African Americans

Caucasian T-score formulas

BVMT Total Recall:

$$[(\text{Total}\text{learning}\text{scaled}\text{score}-(0.2589\ast (\text{edu}-14.11)+(-0.0515)\ast (\text{age}-37.62)+0.9276\ast \text{sex}+10.0712))/2.8912]\ast 10+50$$

BVMT Delayed Recall

$$[(\text{Delayed}\text{recall}\text{scaled}\text{score}-(0.2084\ast (\text{edu}-14.11)+(-0.0286)\ast (\text{age}-37.62)++10.3007))/2.6989]\ast 10+50$$

HVLT Total Recall

$$[(\text{Total}\text{learning}\text{scaled}\text{score}-(0.5225\ast (\text{edu}-14.11)+1.0035\ast \text{gender}+10.1738))/2.5927]\ast 10+50$$

HVLT Delayed Recall

$$[(\text{Delayed}\text{recall}\text{scaled}\text{score}-(0.5414\ast (\text{edu}-14.11)+1.7324\ast \text{genter}+9.9285))/2.6989]\ast 10+50$$

Stroop Word Reading

$$[(\text{Word}\text{reading}\text{scaled}\text{score}-(0.0819\ast (\text{edu}-13.92)+0.0038\ast (\text{age}-36.17)+(-0.4022)\ast \text{gender}+10.2102)-(-0.00003))/2.9435]\ast 10+50$$

Stroop Color Naming

$$[(\text{Color}\text{naming}\text{scaled}\text{score}-(0.0941\ast (\text{edu}-13.92)+10.4444))/2.8101]\ast 10+50$$

Stroop Color-Word

$$[(\text{Color}-\text{word}\text{scaled}\text{score}-(0.2479\ast (\text{edu}-13.92)+(-0.0828)\ast (\text{age}-36.22)++10.3968))/2.6002]\ast 10+50$$

Stroop Interference

$$[(\text{Interference}\text{scaled}\text{score}-(0.2346\ast (\text{edu}-13.92)+(-0.0762)\ast (\text{age}-36.22)+0.7465\ast \text{sex}+9.9952))/2.6342]\ast 10+50$$

WCST-64 Total Errors

$$[(\text{Total}\text{Errors}\text{Scaled}\text{score}-(0.3187\ast (\text{edu}-14.13)+(-0.01)\ast (\text{age}-37.37)+0.1608\ast \text{gender}+10.4049)-(-0.0017))/2.674]\ast 10+50$$

WCST-64 Perseverative Errors

$$[(\text{Perseverative}\text{Errors}\text{Scaled}\text{Score}-(0.2357\ast (\text{edu}-14.13)+(-0.0941)\ast (\text{age}-37.37)+0.0341\ast \text{gender}+10.33)-(-0.0012))/2.5506]\ast 10+50$$

WCST-64 Conceptual Level Responses

$$[(\text{Conceptual}\text{Level}\text{Responses}\text{scaled}\text{score}-(0.3223\ast (\text{edu}-14.13)+(-0.0941)\ast (\text{age}-37.37)+0.0577\ast \text{gender}+10.4292)-(-0.0016))/2.568]\ast 10+50$$

African-American T-score formulas

BVMT Total Recall:

$$[(\text{Total}\text{learning}\text{scaled}\text{score}-(0.2834\ast (\text{edu}-13.86)+(-0.1125)\ast (\text{age}-40.63)+1.0394\ast \text{sex}+8.0679))/2.5701]\ast 10+50$$

BVMT Delayed Recall

$$[(\text{Delayed}\text{recall}\text{scaled}\text{score}-(0.2267\ast (\text{edu}-13.86)+(-\mathrm{0.12.62})\ast (\text{age}-40.63)+0.8593\ast \text{sex}+7.691))/2.5197]\ast 10+50$$

HVLT Total Recall

$$[(\text{Total}\text{learning}\text{scaled}\text{score}-(0.2917\ast (\text{edu}-13.86)+(-0.0644)\ast (\text{age}-40.63)+1.1462\ast \text{sex}+8.3063))/2.8333]\ast 10+50$$

HVLT Delayed Recall

$$[(\text{Delayed}\text{recall}\text{scaled}\text{score}-(0.3986\ast (\text{edu}-13.86)+(-0.0733)\ast (\text{age}-40.63)+.9145\ast \text{sex}+8.2753))/3.1354]\ast 10+50$$

Stroop Word Reading

$$[(\text{Word}\text{reading}\text{scaled}\text{score}-(0.3557\ast (\text{edu}-13.92)+(-0.0866\ast (\text{age}-40.67)+1.2315\ast \text{sex}+8.3263)-(0.00095))/2.8127]\ast 10+50$$

Stroop Color Naming

$$[(\text{Color}\text{namimg}\text{scaled}\text{score}-(0.3102\ast (\text{edu}-13.92)+(-0.1006)\ast (\text{age}-40.67)+1.4915\ast \text{sex}+8.2672))/2.6643]\ast 10+50$$

Stroop Color-Word.

$$[(\text{Color}-\text{Word}\text{scaled}\text{score}-(0.2363\ast (\text{edu}-13.94)+(-0.1219)\ast (\text{age}-40.67)+1.9479\ast \text{sex}+7.449))/2.2658]\ast 10+50$$

Stroop Interference.

$$[(\text{Interference}\text{scaled}\text{score}-((-0.0303)\ast (\text{age}-40.67)+(1.4688\ast \text{sex}+8.1343))/2.663]\ast 10+50$$

WCST-64 Total Errors

$$[(\text{Total}\text{Errors}\text{Scaled}\text{score}-(0.3321\ast (\text{edu}-13.97)+(-0.0838)\ast (\text{age}-40.7)+0.3215\ast \text{sex}+8.1621)-(-0.0006))/2.7911]\ast 10+50$$

WCST-64 Perseverative Errors

$$[(\text{Perseverative}\text{Errors}\text{Scaled}\text{Score}-(0.3599\ast (\text{edu}-13.97)+(-0.0776)\ast (\text{age}-40.7)+(-0.1093)\ast \text{sex}+8.5524)-(0.0006))/3.0124]\ast 10+50$$

WCST-64 Conceptual Level Responses

$$[(\text{Conceptual}\text{Level}\text{Response}\text{scaled}\text{score}-(0.4002\ast (\text{edu}-13.97)+(-0.0874)\ast (\text{age}-40.7)+0.2546\ast \text{gender}+8.2248)-(0.0006))/2.5887]\ast 10+50$$

Notes

Gender

Male = 0

Female = 1

Education

Years of education were determined using a previously defined and standardized procedure where education level ranges from 1–20 based on number of years of schooling completed (Heaton, et al., 2004).