| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
| ||||||||||||||||||||||||||||||||
RESEARCH ARTICLE |
1 Department of Kinesiology and Community Health, University of Illinois at Urbana-Champaign.
2 Department of Kinesiology, The Pennsylvania State University, University Park.
Address correspondence to Jacob J. Sosnoff, Department of Kinesiology and Community Health, 906 S. Goodwin Avenue, 207 Freer Hall, Urbana, IL 61801. E-mail: jsosnoff{at}uiuc.edu
 |
Abstract |
|---|
 TOP Abstract MethodsResultsDiscussionReferences |
|---|
Aprominent theory seeking to explain the mechanism(s) for normal age-related motor deficits is the neural noise hypothesis (Crossman & Szafan, 1956
; Gregory, 1957; Li, Huxhold, & Schmiedek, 2004; Welford, 1958, 1981). This hypothesis proposes that, with advanced age, there is an increase in random neural fluctuations (i.e., neural noise) that interferes with the transmission of information within the central nervous system, leading to a decrement in performance. The neural noise theory of aging holds relevance to the recent interest in within-participant variability in cognitive aging (Hultsch & MacDonald, 2004; Martin & Hofer, 2004; Nesselroade & Salthouse, 2004). If noise within the sensorimotor system is driving age-related performance decrements, then performance variability across tasks should be positively related and the strength of this relationship should be greater in more closely related tasks.
Evidence for this role of neural noise is found in researchers' observations that individuals with poor cognitive function are more variable and the degree of variability in one cognitive domain is related to variability in other domains (Hultsch, MacDonald, Hunter, Levy-Bencheton, & Strauss, 2000). Researchers have also found that increases in intra-individual variability in reaction-time tasks are indicative of variability in other cognitive tasks (Strauss, MacDonald, Hunter, Moll, & Hultsch, 2002). This generalization in the variability across tasks has been interpreted as evidence that measures of behavioral variability are capable of serving as an indicator of neural variability (i.e., noise). In addition, because individuals with cognitive impairments (such as mental retardation or traumatic brain injury) have greater variability on reaction-time tasks, it has been proposed that variability on such tasks is an index of neurological health (Hultsch & MacDonald, 2004).
In contrast to the work done in cognitive aging, examinations of variability on movement tasks have not found any significant relation or generalizability in the variability across tasks (Beggs, Sakstein, & Howarth, 1974; Kail, 1997; Robertson et al., 1999; Zelaznik, Spencer, & Doffin, 2000). Partly on the basis of this lack of generalizability of the amount of intra-individual variability, Newell and Vaillancourt (2001) have theorized that motor variability is specific in part to the intrinsic dynamics of the task.
The vast majority of studies on generalizability of variability in the cognitive and motor domains have focused on distributional characteristics of performance such as the mean and the standard deviation. The use of distributional measures of performance results from the assumption that the underlying mechanism contributing to performance variability is additive white Gaussian noise that is random with independence between successive values (Newell & Slifkin, 1998; Welford, 1958, 1981). However, there is a growing body of literature reporting that these traditional distributional measures do not sufficiently characterize performance (Gilden, Thornton, & Mallon, 1995; Newell & Slifkin; Riley & Turvey, 2002; Ward, 2002), because they do not account for the time-sequential and frequency-dependent properties of performance. Within the cognitive domain, Gilden (2001) has suggested that the time-sequential properties account for a considerable amount of variance in psychomotor tasks and, in some cases, even more than standard cognitive psychological distributional measures.
Theories, such as the neural noise theory, that are based on the driving of performance variability by random noise are unable to account for the temporal structure in performance variability (Newell, Deutsch, Sosnoff, & Mayer-Kress, 2006). Lipsitz and Goldberger (1992) proposed that losses in function in advanced age are characterized by a generalized loss of complexity (i.e., an increase in time-sequential and frequency-dependent structure). Although there are a number of experimental outcomes that are consistent with the loss-of-complexity hypothesis (Goldberger et al., 2002; Kryrias, 2003; Newell, Vaillancourt, & Sosnoff, 2006), the vast majority of studies have focused on the output of a single process (cf. Goldberger et al.). Therefore, there is limited direct evidence of the generalized reduction in complexity as a function of age. Furthermore, there is a theoretical basis and experimental evidence for the idea that the change in complexity with aging is task dependent (Vaillancourt & Newell, 2002, 2003).
In this investigation we examine the generalizability of the amount and structure of variability within an individual across a small subset of motor tasks as a function of the aging process. The motor tasks we selected all involve the same effector (index finger) acting either alone or in conjunction with other digits. Our choice of a common effector across tasks was to enhance the commonality of the sensorimotor mechanisms involved in motor output and thus the potential of the generalizability across tasks of the perceptual-motor variability. On the basis of our previous experiments with single tasks, we hypothesized that the variability would be made up by a range of frequency contributions that were adaptive to task and age influences, and a white noise contribution that would be a very small part of the variability. In line with the neural noise hypothesis, we predicted that there would be generalizability in the amount of perceptual-motor variability across tasks and that it would be greatest in the older adults. Congruent with the loss-of-complexity hypothesis, we also hypothesized that there would be a significant association across tasks in the time- and frequency-dependent structural measures of performance variability that increases with the aging process.
|
METHODS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
Tasks and Procedures
To examine the intra-individual variability as a function of age, we had participants complete four motor tasks of increasing difficulty as part of a larger project examining age differences in visual-motor processing (See Sosnoff & Newell, 2006). The four tasks were a finger postural tremor task, a single-digit isometric force task, and two- and three-digit isometric grip tasks. Participants completed the two- and three-digit grasping task in the same experimental session as the postural tremor task. In order to minimize the influence of fatigue, the tremor task was always completed first. We had the single-digit isometric force task performed during separate experimental sessions. We randomized the order of these two experimental sessions across subjects.
We chose the tasks utilized for this experiment for several reasons. First, the tasks all require control of the index finger and thus have some common effector demands. Second, these tasks quickly and efficiently yield large data sets, which are needed for the calculation of the time-sequential properties. Third, the precise production of force is a hallmark feature of the control of movement (Carlton & Newell, 1993). Fourth, the difficulty of the grasping task scales to the number of digits involved (Sharp & Newell, 2000). In short, in multidigit tasks, participants have to determine how much force to produce with each digit in order to produce the appropriate total amount of force. Lastly, all of the tasks require steady-state output and thus are indicative of processing robustness (Li et al., 2004).
Postural tremor task
Tremor is the involuntary rhythmic oscillations of a limb resulting from descending neural inputs and mechanical factors. It serves as an index of the health of the sensorimotor system. We used the traditional approach to record and quantify index finger postural tremor (Elble & Koller, 1990). Participants sat in a chair with their dominant forearm prone on a table top and their index finger extended, while their remaining digits formed a fist. An accelerometer (model T45-10, Coulbourn Instruments, Allentown, PA) capable of quantifying movement acceleration was placed on the tip of the index finger. We converted the analog output of the accelerometer to a digital signal via a 16-bit analog-to-digital card at 140 Hz and saved it on the hard drive of a computer. Participants completed two 10-s trials and were instructed to minimize the movement of their index finger during each trial.
Index finger isometric force task
Participants were seated in a chair facing a 17-in. (43.18 cm) video monitor (CTX International, California), with their dominant hand placed in a prone position in an apparatus designed to isolate abduction force of the index finger (Sosnoff & Newell, 2006). The apparatus was placed on a table 75 cm above the ground. The movement of the middle finger was restricted, while the thumb was free to move. Orthogonal to the index finger and 36 cm from the subject's midline was an Eltran EL-500 load cell (diameter = 1.27 cm). The load cell measured compressive force produced by the index finger and had a very low level of background noise of 3.5 µV. The analog force signal was sampled at 140 Hz by a 16-bit analog-to-digital converter. On the monitor, participants viewed a red target line, which spanned the width of the monitor, and a series of illuminated yellow pixels; both were on a black background. The red target line corresponded to 5% of a participant's maximal force output and the yellow pixels depicted the force trajectory produced by the participant. Participants completed three 30-s trials. They were instructed to minimize deviations from the target. We took the data for the index finger force production from a larger data set examining age-related changes in visual-motor processing (Sosnoff & Newell).
Two- and three-digit isometric force output
In this task, participants produced force with either their index finger opposing the thumb (two-digit grip) or their index and middle finger opposing the thumb (three-digit grip; see Sosnoff, Jordan, & Newell, 2005). Participants sat on a chair approximately 70 cm in front of a computer monitor, facing the force apparatus with their feet flat on the floor. They viewed their total force output as well as a target line on a computer monitor. The target line corresponded to 5% of their maximal force output. They were instructed to minimize deviations from the target. Participants completed three 20-s trials.
We had the force apparatus oriented in such a way that the thumb load cell faced the participant and the index finger and middle finger load cell faced away from the participant. We used a separate apparatus for each grip configuration. The first apparatus consisted of two unidirectional load cell force transducers (ELFS-B3, range 73.11 N, sensitivity 28.51 mV/N, and ELF-500, range 73.11 N, sensitivity 29.10 mV/N; Entran Devices, Inc., Fairfield, NJ), which were mounted on an aluminum frame. The second apparatus consisted of three of the same type of load cells. We collected force data at 140 Hz, amplified it with a Coulburn amplifier (Model S72-25, Coulburn Instruments, Allentown, PA), digitized it with a 16-bit analog-to-digital converter, and stored it on the hard drive of a computer.
Data Analysis
We utilized the relative task error measured by the coefficient of variation (CV, where CV = SD/M) as a measure of task performance (i.e., within-subject variability) in the force tasks, because the force targets were relative to individual maximum force output for a given task. In order to be consistent across tasks, we also used the CV of the rectified tremor signal to index relative task error. We indexed the time-sequential structure of the motor output within a subject with approximate entropy, or ApEn (Pincus, 1991). ApEn yields a single value that quantifies the time-sequential structure of a time series. A highly structured signal such as an ideal sinewave would have an ApEn value approaching zero, whereas a random, nonstructured time series (white noise) would have a value close to 2. Increases in ApEn have been interpreted as an increase in the signal's time-dependent structure (Pincus). It is important to note that we found ApEn to be reliable across trials, with intertrial correlation coefficients ranging from 0.7 to 0.8.
We also examined the structural properties of perceptual-motor variability within an individual in the frequency domain. We used fast Fourier transform analysis to decompose the motor signal into its frequency components. The slope of the log-log frequency profile can be used to index a system's structure (Basingthwaighte, Liebovitch, & West, 1994), with a steeper slope indicating an increased frequency-dependent structure. A completely random signal would have a spectral slope of zero (i.e., white Gaussian noise), whereas a highly structured signal would have an increasing negative slope. To quantify the influence of age on the structure of the perceptual-motor output, we calculated a least square regression over the 0- to 12-Hz region of the log-log power spectrum for each individual in each of the four motor tasks. This region of activity accounts for the majority of power in the motor signal in the force tasks. We performed the analysis on the force data by using Welch's average periodogram method (Mathworks, 1996). We divided the data into 256 data point sections and then linearly detrended it. We calculated the discrete Fourier transform function for each section and then averaged across sections. The frequency resolution of the resulting power spectrum was 0.54 Hz for all tasks. It is important to note that the calculation of the spectral slope is only valid in signals that have a proportional decrease in power as a function of frequency. Consequently, we did not calculate the spectral slope on the tremor data. It is also important to note that we found the spectral slope to be reliable across trials, with intertrial correlation coefficients ranging from 0.7 to 0.8.
Statistical analysis
We performed one-way analyses of variance, with age as the main factor, on within-participant task variability measures, ApEn values, and spectral slope values. When it was relevant, we used Tukey's honestly significant difference test to determine the specific effects contributing to the general analysis of variance. To examine the amount and structure of variability within an individual across tasks as a function of age, we calculated Pearson correlation coefficients for each unique pairing of task variability, ApEn values, and spectral slope values. We evaluated all statistics as significant when there was a less than 5% chance of making a Type I error (p <.05). We completed statistical analyses by using a statistical package (SPSS version 12).
|
RESULTS |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
2 =.10; two-digit force output, F(2, 45) = 1.7, p <.05, 2 =.08; and three-digit force output, F(2, 45) = 3.4, p <.05, 2 =.14. A post hoc analysis revealed that both the 60- and 70-year-old groups (0.117 and 0.082, respectively) were more variable in index finger force output than was the 20-year-old group (0.061). In the two-digit grip, only the 60-year-old group was more variable than the 20-year-old group (0.083 vs 0.055, respectively). In the three-digit grip, the 70-year-old group was more variable than the 20-year-old group (0.096 vs 0.058, respectively).
|
2 =.15. A post hoc analysis revealed that, in the single-digit task, the young group (0.406) had greater ApEn than did the 70-year-old group (0.254).
Because postural finger tremor does not have a proportional decrease in power as a function of frequency (Elble & Koller, 1990), the calculation of spectral slope for such a system is not informative (Basingthwaighte et al., 1994). Consequently, we calculated the spectral slope as a function of age group only on the force production tasks (Table 1). Overall, we found that younger adults had a less steep slope compared with the older age groups in all three tasks: single-digit task, F(2, 45) = 3.11, p <.05, 2 =.11; two-digit task, F(2, 45) = 3.45, p <.05, 2 =.15; three-digit task, F(2, 45) = 3.45, p <.05, 2 =.15. In the single-digit force task, the 20-year-old group (–1.53) had a less steep slope than did the 70-year-old group (–1.93). In the two- and three-digit tasks, the 20-year-old group (–1.67 and –1.72, respectively) had a less steep slope than did the 70-year-old group (–2.11 and –2.06, respectively). In addition, in the three-digit force task, we found the 60-year-old group (–1.64) to have a less steep slope than the 70-year-old group.
Intra-individual Variability Across Perceptual-Motor Tasks
To examine the amount of intra-individual variability in the magnitude and structure of perceptual-motor variability across these four motor tasks, we conducted correlational analyses with age group controlled for. We found a significant positive correlation for task variability between the two-digit and three-digit force task, r(45) = 0.696, p <.05 (see Table 2). There was no significant relationship between any of the other tasks.
|
The spectral slope correlations revealed results congruent with those reported for ApEn (See Table 2). Overall, there was a significant relationship between spectral slope in the single-digit and two-digit tasks, r(45) = 0.619, p <.05, the single-digit and three-digit tasks, r(45) = 0.539, p <.05, and the two- and three-digit tasks, r(45) = 0.624, p <.05.
Age Differences in Intra-individual Variability
Once it was established that there were significant relationships between the magnitude and structure of motor variability across the tasks studied, we found it of interest to examine if aging influenced these relationships. Consequently, within each of the three age groups we calculated Pearson correlations between measures of task performance as well as measures of the structure of motor variability across the four motor tasks.
When broken down between age groups, the relationship between the amounts of variability across tasks was minimized. In fact, there was no significant relationship between any of the tasks across the age groups when there was less than a 5% chance of making a Type I error. However, we did observe a significant relationship in the 70-year-old age group between CV in the tremor and single-digit force tasks, r(13) = 0.457, p <.10, the single-digit and three-digit force tasks, r(13) = 0.485, p <.10, and the two-digit and three-digit force tasks, r(13) = 0.481, p <.10 (see Table 3), when we relaxed the value to p =.10.
|
|
|
|
DISCUSSION |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
), and they provide challenges to theories of interindividual variability that do not account for the sequential structure of perceptual-motor performance. The findings highlight the importance of structural as opposed to only magnitude measures of variability in investigations of variability and aging.
Age Differences in Perceptual-Motor Variability
Not surprisingly, researchers have found older adults to be more variable than younger adults in their perceptual-motor performance (Enoka et al., 2003; Welford, 1981). Researchers have also found older adults to have greater frequency- and time-dependent structure in their performance variability (Lipsitz & Goldberger, 1992; Vaillancourt & Newell, 2002, 2003; Newell, Vaillancourt, et al., 2006). Although we found age-related changes in performance in the three force tasks, there was no effect for age in the postural tremor task. This discrepancy leads us to suggest that age differences in perceptual-motor performance are not a result of fundamental age-related decrements in the health of the sensorimotor system, but rather age-related differences in task-dependent control processes.
Neural Noise Theory of Aging
If neural noise is driving perceptual-motor performance decrements in old age, as proposed by the neural noise theory of aging, it is logical to assume that decrements will be consistent across tasks (Crossman & Szafan, 1956; Welford, 1958, 1981). We found a significant relationship between task performance between the two-digit and three-digit tasks in the oldest age group (at a p <.10 level), but the amount of common variance accounted for was very low. The low level of association between the amount of task variability and the very low levels of white noise in the respective time series do not support the strong tenets of the neural noise hypothesis of aging and variability. The minimal relationship of performance across tasks is congruent with the lack of evidence for a common individual degree of variability within the motor domain (Beggs et al., 1974; Kail, 1997; Robertson et al., 1999; Zelaznik et al., 2000). Our findings are much more consistent with the proposition that variability is specific to the intrinsic dynamics of the task that constrain motor output (Newell & Vaillancourt, 2001). That is to say, the observed behavior is driven by the interaction of the environmental, organismic, and task demands (Newell, 1986), a view that is consistent with more recent interpretations of embodied cognition (Clark, 1997).
Researchers have maintained that, in advanced age, individuals are less capable of adapting their output to the dynamics of the task (Newell, Vaillancourt, et al., 2006; Vaillancourt & Newell, 2002). The stronger relationship between structural measures across tasks in the older age groups is consistent with this proposition. The increased strength in the relationship of complexity measures across tasks within old age groups is also congruent with the notion of dedifferentiation with aging (Li, 2002; Baltes, Cornelius, Spiro, Nesselroade, & Willis, 1980). Thus, it raises the possibility that there is dedifferentiation in perceptual-motor function with advanced age.
It is important to note that even if the relationship between performances across tasks was stronger, it in and of itself would not provide support for the neural noise hypothesis of aging. Although the relationship between performances across tasks has previously been interpreted as evidence in support of the neural noise hypothesis (cf. Li, Aggen, Nesselroade, & Baltes, 2001), there is a growing amount of evidence that measures of performance such as standard deviation and coefficient of variation are not a direct measure of noise (Newell, Deutsch, et al., 2006; Riley & Turvey, 2002). In order for one to support the neural noise hypothesis directly, one has to quantify the amount of noise as a function of age directly, rather than taking the standard deviation of task output as a representative measure (See Sosnoff & Newell, 2005).
The Structure of Perceptual-Motor Variability
There is growing evidence that it is not the amount of white Gaussian noise that drives age-related increases in perceptual-motor variability, but rather changes in the time and frequency structure of motor variability (Deutsch & Newell, 2003; Newell, Deutsch, et al., 2006; Vaillancourt, Sosnoff & Newell, 2004). Although it is possible that increases in neural noise lead to a change in the structure of perceptual-motor output, to date there is minimal evidence supporting this view (Newell, Deutsch, et al.). The concept that there are changes in the structure of motor variability with age is theoretically linked to the dynamical view of aging (Glass & Mackey, 1988; Yates, 1988). This perspective maintains that deficits associated with advanced age are due to temporal aberrations in control processes.
For instance, Lipsitz and Goldberger (1992) proposed that behavioral and physiological decrements caused by aging are a result of systematic decreases in physiological complexity. There is a considerable amount of evidence from physiological and behavioral systems that is consistent with the loss-of-complexity theory (Goldberger et al., 2002). However, the overwhelming majority of the support for the loss-of-complexity theory is based on the output of individuals during a single task.
By examining the structure of motor variability across several tasks, the current investigation provides (to our knowledge, for the first time) evidence for a systematic and general decline in perceptual-motor complexity with age. It is important, that the positive relationship between perceptual-motor structure (i.e., ApEn and spectral slope) was found between all the force tasks and that the strength of this relationship was found to increase in the older age groups.
The current investigation demonstrates the importance of examining the structure of variability across tasks as a function of age. The validity of examining intra-individual variability within the cognitive domain has recently been brought to the forefront in the cognitive aging domain (Martin & Hofer, 2004). The majority of studies utilize measures relating to the magnitude of performance that do not take into account time-sequential or frequency properties (Newell & Slifkin, 1998). However, there have been several investigations that have applied time-sequential indices to cognitive processes (e.g., Gilden et al., 1995; Wagenmachers, Farrell, & Ratcliff, 2004; Ward, 2002). Gilden (2001) has suggested that these measures are able to account for a considerable proportion of variance in psychomotor tasks and in some cases an even greater amount than standard cognitive psychology measures. The findings of the current investigation show that these structural measures of variability will add to the growing knowledge base involving the aging process and variability.
It is possible that the lack of significant relationship between the magnitude of variability across these perceptual-motor tasks is a result of relatively small sample size (N = 48; 16 per age group). For instance, in a recent study, Nesselroade and Salthouse (2004) showed in a much larger sample (N = 204) that the magnitude of performance on a tracking task in a given session was related to variability across testing sessions and that both indices of variability were linked to cognitive function. Nevertheless, the sample size of the current investigation was large enough to detect age differences in both the magnitude and structure of within-subject perceptual-motor variability.
Conclusion
The current investigation reveals that the magnitude of perceptual-motor variability was only minimally related across tasks, and thus it offers no support for the neural noise hypothesis of aging. In contrast, the structure of perceptual-motor variability was found to be more highly related across the tasks, and this relationship was greatest in the older age group. This study provides preliminary evidence for a systematic general decline in complexity (i.e., increased structure) within the perceptual-motor system with advanced age (Lipsitz & Goldberger, 1992). It is proposed that, with advanced age, individuals are generally less able to adapt the structure (i.e., complexity) of their perceptual-motor output to the various demands of the task (Newell, Vaillancourt, et al., 2006; Vaillancourt & Newell, 2002).
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Received for publication October 19, 2005. Accepted for publication March 23, 2006.
|
References |
|---|
|
TOPAbstractMethodsResultsDiscussionReferences |
|---|
noise in human cognition. Psychonomic Bulletin and Review, 11,579-615.This article has been cited by other articles:
 |
|
 |
|
  J. J. Sosnoff and K. M. Newell Age-Related Loss of Adaptability to Fast Time Scales in Motor Variability J. Gerontol. B. Psychol. Sci. Soc. Sci., November 1, 2008; 63(6): P344 - P352. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||
| HOME | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|---|
