Despite the widespread clinical use of CDTs, there are limited published data regarding clock drawing performance in representative community-based samples of older adults (Hubbard et al., 2008). This is particularly true of the CLOX measure, for which cutoffs to determine impairment are based on performance in a sample of 62 college students (Royall et al., 1998). In general, data on representative samples are important for establishing boundaries of normal performance within a population before extrapolating that data to the determination of clinical impairment (see Lezak, Howieson, & Loring, 2004, or Strauss, Sherman, & Spreen, 2006 for detailed discussion on the need for normative data to interpret cognitive performance). One of the problems with using cognitive performance of younger adults to judge impairment in older populations is highlighted by findings from the Seattle Longitudinal Study, one of the few studies that has been able to adequately examine cohort differences in cognitive performance by using a sequential design (Shaie, 1994). Adults from more recently born cohorts show better performance across a number of different cognitive tasks than previously born cohorts (Shaie, 1994). Thus, using performance of younger adults to determine impairment in older adults would be likely to lead to overestimation of cognitive impairment in older people.

Of the few community-based studies describing performance on CDTs in larger samples of older adults, older age and fewer years of education are typically associated with poorer CDT performance (Bozikas, Gaizkoulidou, Hatzigeorgiadou, Karavatos, & Kosmidis, 2008; Ravaglia et al., 2003; von Gunten et al., 2008). Similar associations with age and education have also been found specifically for CLOX performance (Crowe, Clay, Sawyer, Crowther, & Allman, 2008; Schillerstrom et al., 2007). In addition, several studies have highlighted the concern that quantity of educational experience (i.e., years or level of education) fails to capture quality of educational experience, which can vary widely across cohorts, states, and even counties within states (Dotson, Kitner-Triolo, Evans, & Zonderman, 2008; Manly & Echemendia, 2007). This idea of accounting for quality of education is especially important for interpreting performance on cognitive tests among older African Americans, since those who attended segregated schools were more likely to have experienced a poorer quality of education due to factors such as inferior funding, greater number of students per classroom, and shorter length of school year (Manly, Jacobs, Touradji, Small, & Stern, 2002).

Several studies have demonstrated that despite comparable years of education, older African Americans show lower scores than Caucasian older adults on a variety of different standardized neurocognitive tests (Baird, Ford, & Podell, 2007; Boone, Victor, Wen, Razani, & Ponton, 2007; Manly et al., 2002; Pedraza & Mungas, 2008). It is important to note that many of these discrepancies are reduced to non-significance when reading level is used as a covariate (Manly et al., 2002). Reading level is thought to be a proxy measure for educational quality. In analyses specific to CLOX, African American race was found to be associated with poorer CLOX performance, even after adjustment for years of education (Crowe et al., 2008). However, after controlling for reading ability, racial differences on CLOX1 scores were eliminated. Overall, age and reading ability were the two factors that were most strongly related to CLOX1 and CLOX2 performance in multiple regression models including race, gender, urban/rural residence, and level of education (Crowe et al., 2008).

Similar to other cognitive tests, the concern with using CDTs as markers of cognitive dysfunction in older adults is that without a comprehensive understanding of the natural variability of test performance across a variety of older adult populations, there is a potential for over- or underestimation of impairment (Manly, 2008; Pedraza & Mungas, 2008). Pinto and Peters (2009) provided a recent comprehensive literature review on CDTs in clinical screening and found that several CDT versions exhibited good ability to discriminate between moderate/severe dementia and controls, but displayed much more limited sensitivity/specificity values in distinguishing between milder dementia cases and controls. These studies highlight the need for further examination of CDT performance in cognitively healthy older adults.

Table 1 also shows sample characteristics specifically for participants who were and were not included in the normative sample, as well as statistical comparisons of these groups for demographic and cognitive characteristics. Those who were excluded from the normative sample were older and more likely to be African American, male, and from rural counties of Alabama. However, the normative sample still comprised nearly 40% of African Americans. Differences between the excluded and normative samples were also found for education level, reading ability, and cognitive test scores. The normative sample was more highly educated and had better scores on the WRAT-3, MMSE, and CLOX tests.

Age and reading ability groups were based on tertiles. This strategy produced fairly equivalent sample size representation across the age and reading ability groups. The age groupings consisted of participants who were 65–69, 70–74, and 75 or more years old. For reading ability, scores on the WRAT-3 reading subtest were <38, 39–46, and >47 for the three groups. Corresponding reading grade level for these groups was 6th grade or less, 7th grade to high school, and high school to post-high school, respectively.

The central focus of the current study was to present CLOX performance data using age and reading level as the grouping variables of interest. Several recent studies have highlighted the issue that reading ability may be more closely linked to the estimation of quality of education than number of years of education, an issue that may be particularly salient in understanding and interpreting performance on cognitive measures among older African Americans who may have attended segregated schools (Manly et al., 2002). Differences in cognitive test performance among older members of disadvantaged racial groups may not reflect a higher prevalence of cognitive disorders but disparities in access to better quality of education. Our community-based sample of older adults was relatively unique in representing substantial proportions of persons with low reading levels, African American race, and rural residence.

Based on evidence demonstrating a strong association between reading ability and cognitive test performance, other studies are beginning to present normative test data based upon word reading skills (e.g., Hubbard et al., 2008). In light of the increasing number of studies showing a relationship between better reading ability and higher scores on various cognitive tests, as well as our previous findings specifically related to CLOX (Crowe et al., 2008), we presented normative information by age and reading level categories instead of the more traditionally used years of education. However, it should be noted that not all studies have demonstrated an association between reading level and clock drawing performance. For example, one recent study investigating older African American community-dwelling adults reported that clock drawing performance was not related to reading ability (Johnson, Flicker, & Lichtenberg, 2006).

Previous normative CLOX data were presented for a cross-sectional sample of older adults that was 99% Caucasian and living in a continuing care retirement community (Royall, Chiodo, & Polk, 2003). The current study utilized a more representative sample of older adults, and we were able to exclude individuals from the normative sample who showed substantial global cognitive decline. In addition, our study is the first to our awareness to provide normative data for any clock drawing test for persons with reading levels below 7th grade, given that our lowest reading ability category corresponded to a 6th grade or lower reading level. One recent study (Hubbard et al., 2008) published detailed normative clock drawing data for three other clock scoring systems using a sample of 207 African American and Caucasian middle-aged and older adults (ages 55–98). However, the study sample consisted of persons with higher reading ability, with a mean WRAT-3 reading score of 52, in contrast to a mean WRAT-3 reading score of 42 in this normative sample.

Provision of normative data stratified by reading ability level may assist clinicians and researchers begin to discern the impact of educational quality upon cognitive performance. Using previously reported data on the CLOX task, the current sample would have relatively high rates of impairment as defined by CLOX1 scores of <10, which represents performance below the 5th percentile for young adults attending college (Royall et al., 1998). Approximately one-third of our normative sample in the lowest reading ability category would be classified as having impaired executive function using this cutoff. This is despite the fact that many of our participants with the lowest reading and educational levels were excluded from our normative sample based on low MMSE scores at baseline. These findings highlight the necessity to consider educational factors in decisions regarding a diagnosis of cognitive impairment, particularly for persons with lower access to educational opportunities indicated by reading scores lower than expected based on the level of educational attainment. Interestingly, while only 6% of the normative sample in the current study had 0–6 years of education, 33% of our normative sample had reading scores at the 0–6th grade level.

In terms of study limitations, we relied upon WRAT-3 reading scores from the 4-year follow-up home evaluation since reading testing was not conducted at baseline. There is, therefore, a potential bias that reading scores may have changed over the 4-year period from baseline to follow-up. However, prior research on stability of reading ability suggests that reading scores are fairly robust measures over time and are relatively insensitive to dementia until later in the disease stages (Ashendorf et al., 2009; McCaffrey, Duff, & Westervelt, 2000). Another limitation is that dementia was not formally assessed in this study. We excluded from the normative sample any participant with identified clinical diagnosis of dementia at either baseline or the 4-year follow-up home assessment, as well as excluded individuals with either low MMSE scores at baseline (<24) or a four-point or more MMSE decline over the 4-year period. On the other hand, the use of a cutoff on baseline MMSE may have excluded individuals with low scores not due to cognitive decline, especially for those with lower levels of education and reading ability. The fact that 36% of the sample was excluded from the normative data gives us confidence that the remaining participants were cognitively healthy overall. However, it is acknowledged that a portion of the study sample may have been clinically characterized with cognitive impairment (i.e., mild cognitive impairment or early stage dementia; Petersen et al., 2009) if formal clinical assessments had been conducted.

There is some controversy in the field of neuropsychology about the use of norms based on demographic factors. An advantage of norms that take demographic factors into account is enhanced specificity (reduced risk of false-positive diagnoses of cognitive impairment), especially among those with lower levels of education or reading levels (Heaton, Miller, Taylor, & Grant, 2004; Lezak et al., 2004; Marcopulos, Gripshover, Broshek, McLain, & Brashear, 1999; O'Connell & Tuokko, 2010; Strauss et al., 2006). However, data also suggest that removing the effects of risk factors for cognitive impairment (e.g., age and education) reduces sensitivity of a test and increases the risk of a false-negative diagnosis (Fastenau, 1998; Morgan & Caccappolo-van Vliet, 2001; O'Connell & Tuokko, 2010; Reitan & Wolfson, 1995, 2005; Sliwinski, Buschke, Stewart, Masur, & Lipton, 1997). Ultimately, the decision of whether to use adjusted norms should be made after considering the context of the evaluation, including the base rate of the disorder in the subgroup, the sensitivity and specificity of the measure, and costs associated with false-positive and false-negative diagnoses (see Strauss et al., 2006). In the context of a clinical evaluation for dementia, the use of norms appropriate for age and reading level is recommended given the potential high costs of a false diagnosis of dementia.