In the current study, we examined the factor structure of a battery of neuropsychological tests commonly conceptualized as attention measures. Our aims were to assess the applicability of an existing multielement model of attention in a sample of HIV-infected adults, many of whom were drug users, and to determine whether the individual derived factors were associated with a variety of cognitive, behavioral, and virologic variables. This was the first time the model described by Mirsky and colleagues (1991)
has been applied to an HIV+ population.
The factor structure based on our data is very similar to that described by Mirsky and Duncan (2001)
and others (Kelly, 2000
; Kremen et al., 1992
; Mirsky & Duncan, 2001
; Pogge et al., 1994
), in that there were similar loadings of the measures across factors. Five factors resulted from our PCA. The first, similar to the focus/execute
element described by Mirsky, consisted primarily of measures requiring speeded visual processing and some type of cognitive operation, such as form discrimination (Symbol Search), visuomotor integration (Digit Symbol), and response inhibition (Stroop Interference). Whether or not focus
are accurate descriptors for the underlying similarity among these measures is open to debate. In this case, the interpretation difficulty of such a factor is not due to the multifactorial nature of attention, as tasks with specific processing demands can assess specific aspects of attention. Rather, as mentioned before, the difficulty lies in the multifactorial nature of these tests. Without good (or any) control conditions, it can be tricky to ascertain what links a group of test measures. Perhaps visual processing speed
would be a more accurate characterization for this factor, as it was only timed visual measures that comprised this factor. Thus, all measures on this factor shared in common a visual stimulus and administration under the pressure of time. However, CPT reaction time, which also presumably reflects visual processing speed, did not load on this factor. This could be due to the difference in cognitive demands between the CPT and the other measures.
Factor 2, called encode
by Mirsky and others, consisted exclusively of verbal tests. These tests required working-memory ability, as well as maintenance of information in a temporary buffer. Such buffers have been called a phonological loop in the past (Baddeley, 2001
). Therefore, it is conceivable that this factor could be considered one of working memory
and/or basic auditory attention
. Indeed, as shown in , there is a trend among researchers in recent years to include some of the tests that comprised this measure in an “attention/working memory” domain. Thus, according to our results, this domain label may be accurate.
It is important to point out that the inherent characteristics of the tests that comprised Factors 1 and 2 (i.e., visual vs. auditory) may be the primary reason that they appear as different factors. In other words, it may be that the sensory modalities through which these tests are successfully completed are the underlying “factors.” It will be necessary to include cross-modality tests in order to investigate this problem. For example, including a spatial span test, considered to be one of basic attention and working memory, will help determine the nature of these factors.
Factor 3, Mirsky's switch
element, was composed solely of the WCST variables, consistent with the previous studies cited above. Switching is synonymous with alternating attention and requires the ability to disengage from one stimulus or mental set and to reengage in another. However, this is an ability that is arguably necessary for successful performance on measures such as Part B of the Trail Making Test. This measure did not load at all on Factor 3. Therefore, it is possible that this factor is reflecting some other ability inherent to the WCST. The WCST itself is a highly complex test with a number of underlying factors according to a recent study (Greve, Stickle, Love, Bianchini, & Stanford, 2005
). Thus, it will be necessary to include additional, yet more simple, measures of set shifting in future analyses to confirm the validity of this as the switch
Factor 4, equivalent to Mirsky's stabilize
, was comprised of variability and omission variables of the CPT. Based upon the definition provided by Parasuraman and Davies (1984)
, this appears consistent with sustained attention, or vigilance. This element is of special interest, because it is not commonly assessed psychometrically in neuropsychological evaluations per their definition. Furthermore, common everyday tasks (e.g., driving) require vigilance in addition to other aspects of attention. We have recently shown the importance of the CPT in providing additional information regarding attentional functioning in those with HIV (Levine et al., 2006
). Looking specifically at sustained attention among HIV+ individuals, stimulant users were found to have a greater numbers of omission errors and variability in reaction time than had non-drug-users, indicative of impaired sustained attention. No difference was found on a general global neuropsychological ability rating or, importantly, on other tests comprising the attention domain between the groups. That finding underscores the importance of a multifaceted assessment of attention.
Finally, Factor 5 was composed of reaction time and commission errors from the CPT. Not surprisingly, the faster the reaction time in our sample, the greater number of commission errors. This factor was termed sustain
by Mirsky and Duncan (2001)
, and was described as the capacity to maintain a “vigilant attitude” over time. This is differentiated from CPT variables that comprise the stabilize
factor, which indicates “consistency or stability with which a person can respond to a designated target stimulus.” Arguably, these are actually two aspects of sustained attention, or vigilance, as defined by Parasuraman and Davies (1984)
. According to their definition, sustained attention is the ability to maintain a certain level of performance, especially in the ability to detect the occurrence of infrequent or unpredictable events over extended periods of time. Further, demonstration of a vigilance problem requires an interaction among task conditions, such as an incremental decline in response speed or accuracy over time. Therefore, both the stabilize
elements from Mirsky's model (Factors 4 and 5 in our analysis, respectively) may be considered aspects of sustained attention, as both are important for determining a vigilance problem. Alternatively, Factor 5 may also be a reflection of impulsivity, response style (d
′), or simply reaction time.
The factors had modest correlations with demographic and behavioral measures. Verbal IQ, as estimated via a reading task, had the strongest correlation with Factor 2. That verbal IQ would be related to Factor 2 is not unexpected considering that it was composed exclusively of verbal tests (Digit Span, PASAT, and Letter–Number Sequencing). Estimated verbal IQ was also mildly associated with Factors 1 and 4. One functional outcome variable, medication adherence, was associated with Factor 4, considered by the authors to reflect sustained attention. Individuals with greater adherence, expressed as a high percentage of prescribed doses that were taken over the course of a 6-month study according to the MEMS cap data, performed with less variability and fewer omission errors on the CPT. Thus there appears to be a parallel between this laboratory measure of inconsistency and omissions and a real-world measure. Finally, depression was correlated with our Factor 5. Our interpretation of this factor as one of impulsivity may be accurate if one considers a common underlying substrate for impulsivity and some aspects of depression. This has in fact been reported by others from our laboratory. Specifically, it was shown that specific items on the BDI cluster together and covary with frontal/executive cognitive abilities (Castellon et al., 2006
). In addition, Castellon, Hinkin, and Myers (2000)
showed that apathy, a common symptom of depression, is associated with performance on a response inhibition task. Thus, it is conceivable that as depression increased in our sample so did a risky response style, or impulsivity in responding, on the CPT.
There are several limitations to our study that need to be mentioned. First, as in previous studies, two or more variables from a single measure were all that represented a particular factor in some instances. For example, number of errors, number of categories, and number correct from the WCST were the sole variables constituting Factor 3, our equivalent of Mirsky's shift
element. Because variables derived from the same measure tend to be highly correlated, this results in multicollinearity of variables and therefore an artificially inflated correlation between them. In future studies, it would be useful to have at least one additional measure to assess the switch, sustain, and stabilize categories. Second, as with the majority of the replication studies described earlier, we employed PCA, which is generally theoretically sound when used as an exploratory method for elucidating patterns of correlations among a set of variables (Tabachnik & Fidell, 1996
). However, the degree to which the derived factors represent latent underlying variables or true constructs is uncertain with PCA alone. Further, because we sought to assess the validity of an existing model, it can be argued that a confirmatory approach would have been warranted. PCA is not recommended for use as a confirmatory tool, as there are strict requirements for the data, including very large sample size, use of “marker” variables, and adequate spread in scores on the variables of interest (Tabachnik & Fidell, 1996
). Taking a different tact, Strauss et al. (2000)
used structural equation modeling in order to as a confirmatory factor analytic approach in their attempt to validate Mirsky's model. Using structural equation modeling and a Mirsky's original four-factor solution, they failed to replicate the earlier findings. However, structural equation modeling also has significant theoretical limitations, and our decision to employ PCA was based on our goal of replicating previous investigations of Mirsky's model. Clearly, continued analysis using a variety of methods and neuropsychological measures are necessary to assess the validity of our interpretation of Mirsky's factors. While data reduction strategies such as PCA are useful as an initial step in uncovering behavioral constructs from performance across a myriad of tests, additional strategies are required to determine whether those factors represent actual endophenotypes with specific anatomical or neural systems. Both Mirsky et al. (1991)
and Posner and Dehaene (1994)
have suggested that different neural substrates underlie the various elements of their theoretical models. Some support has been established for Posner's model, included functional imaging and genetic association studies (Fan & Posner, 2004
; Fan, Wu, Fossella, & Posner, 2001
; Fossella et al., 2002
). However, no such data are available for models such as Mirsky's. This may be due to the psychometric (multifactorial) nature of Mirsky's model. The research that has grown from Posner's model has generally relied upon a simple visual attention paradigm called the Attention Network Test (Fan et al., 2001
), perhaps making physiological and genetic associations more feasible. Therefore, additional effort is required to empirically verify the elements found in this and previous studies of attentional elements derived from traditional neuropsychological tests.