The
“vigilance decrement” is one of the most consistent findings in studies and
experiments on vigilance. First articulated by Mackworth in 1948, this
decrement describes the marked reduction in performance observed in (correct)
signal detection as a function of time. Several theories, such as signal
detection theory, have arisen to explain this decrement, and by the late 1970s,
two primary factors emerged as key players in this vigilance decrement:
sensitivity to the presented signals (d’) and response criteria (beta). In
short, some groups attributed this decrement to a possible decrease in
perceptual sensitivity as a function of time while others (such as Broadbent
and Gregory, 1965) reported subjects’ decrease in confidence as a key cause. While
researchers agreed that both parameters were important, the relative and
dynamic contributions of each factor remained inconclusive. Many questions
still remained. Was the vigilance decrement caused by different factors in
different types of studies? Was it the same in auditory vigilance tasks? Are
perceptual sensitivity and response criterion both involved in vigilance
decrements across all studies? Enter Parasuraman Vigilance Taxonomy.
First
described by Raja
Parasuraman in 1979, Parasuraman Vigilance Taxonomy (PVT, as it will be
referred to henceforth) provides a framework upon which to characterize the
vigilance decrement as a function of the type of task. Specifically, it divides
tasks into two groups: those that loads memory and those that don’t. PVT holds
that different factors are responsible for the vigilance decrement in these two
different types of tasks. In this post, we explore the Parasuraman Vigilance
Taxonomy in detail by first providing a comprehensive description of it and
then citing evidence in support of the taxonomy. Although most of the evidence
presented in the papers for today are in favor of the PVT, it is still
extremely important to consider the contrary evidence, and this is exactly what
we will do. On a deeper level, we will perform a comparative analysis in which
we examine proposed revisions to PVT, such as additions of new dimensions, in
particular the sensory-cognitive distinction described by See et. al. Furthermore, We will also
examine the strengths and weaknesses of the taxonomy. Subsequently, we will
perform a deeper analysis of the studies performed and raise questions
concerning the experimental design and result interpretation. Lastly, we
conclude with PVT’s links to neuroscience and possible future directions.
DESCRIPTION OF PVT: WHAT IS IT?
The Parasuraman Vigilance Taxonomy
refers to a framework for distinguishing between the contribution of perceptual
sensitivity and response criteria in the vigilance decrement. Specifically, as
Raja Parasuraman detailed in his 1979 paper, this taxonomy holds that the
vigilance decrement results from a decrease in perceptual sensitivity only during tasks that loads memory and have rapid event rates. Thus, tasks such as
these, termed successive-discrimination
tasks by Parasuraman, require the subject to compare the presented signal with
previous signals represented in memory when they make a decision. As such, they
require the subject to remember the intensity of previous signals in order to
make a comparison, and thus impose a type of cognitive load on the subject. In
the second type of task, termed simultaneous
discrimination by Parasuraman, refers to tasks in which the signal and
background events are presented together. Here, the subject is asked to
discriminate between the signal and non-signal without needing to remember
previous signals and non-signals. As such, these tasks do not load memory.
According to PVT, the vigilance decrement in simultaneous discrimination tasks results from changes in the subject’s response criteria.
EVIDENCE: SUPPORATIVE AND OTHERWISE;
POSSIBLE REVISIONS TO PVT
Much evidence
exists to support the Parasuraman Vigilance Taxonomy. Here, we review three
studies that investigated this topic (Parasuraman, 1979; See et. al., 1995; Warm
et. al., 2008), with particular
focus on their experimental design, results, and how these results provide
evidence in favor of the Parasuraman Taxonomy. Since most of the later
commentary will concern the Parasuraman paper, we will provide an in-depth
review of his methods and results and a brief overview of the other papers,
which will be included in the comparative analysis section.
Summary of Parasuraman,
1979
Methods and Experimental Design: Subjects
were asked to perform two types of tasks: successive-discrimination and
simultaneous-discrimination tasks (defined above). In the first experiment,
subjects performed an auditory
vigilance task. In the successive-discrimination group, subjects were asked to
distinguish between an intermittent 1000Hz tone (non-signal) and a periodic
2.1-dB increase in that tone (the signal). On the other hand, subjects in the
simultaneous-discrimination group were asked to detect a 1000-Hz tone within an intermittent noise burst. Low
and high event rates (15 events/min and 30 events/min, respectively) were
tested for each group and conditional target probability was controlled for
during each task. In a second experiment, three visual vigilance tasks at a high event rate were compared. In the
successive-discrimination condition, the signal was a decrease in the intensity of a flashing light. In the
simultaneous-discrimination condition, two sources of light were present, and
the signal was defined as a decrease in intensity in one of the light sources.
A second simultaneous discrimination task defined the target as a small circle
appearing at the center of one of the two light sources.
Results/Evidence: In the first (auditory) experiment,
Parasuraman found different results for the successive-discrimination and
simultaneous-discrimination groups. Specifically, subjects in the successive-discrimination
group exhibited a “progressive reduction
in detection ability in consecutive time blocks” at high event rate, as indicated by receiver operating characteristic
plots of the data. This reduction was absent
during the low event rate. For
the simultaneous-discrimination group, however, the data points clustered
around a line of fixed detectability for both the low and high event rates, a
result that indicated that the vigilance decrement observed in this condition
was not due to changes in signal detectability. These results on decreases in
performance in high-event rate successive-discrimination tasks (which load
memory) suggest that perceptual
sensitivity contributes to the vigilance decrement in tasks that load memory
and occur at high event rates, a postulate of the Parasuraman taxonomy.
Furthermore, the finding that event rate has no impact in simultaneous-discrimination
tasks suggest that the vigilance decrement in these tasks could be attributed
to response criterion, another tenant of the Parasuraman taxonomy.
COMPARATIVE ANALYSIS: Proposed Revisions
and Advancing Vigilance Research
The
authors for this week all addressed the PVT in their studies; however, they had
different approaches and conclusions. Here, we perform a comparative analysis
of their findings and inferences. We begin with an examination of the proposed
revisions by See et. al. and then
explore the suggestions for advancing vigilance research posed by Warm et. al.
See
and his research group performed a meta-analysis of 42 vigilance studies in
their effort to refine the PVT by identifying other task parameters that may
contribute to the sensitivity decrement. Although this investigation provided
evidence for Parasuraman’s finding that the decrement is substantial and is dependent on the type of discrimination
and event rate, it also suggested revisions to the existing PVT. In
particular, See et. al. found that
vigilance studies should distinguish between sensory and cognitive tasks. Here,
we will examine See et. al.’s study,
with a particular focus on how their results support and question the
original PVT. Lastly, we look at the proposed revisions proposed in the paper.
Although
much evidence exists to support the PVT, See notes that “additional
investigation have indicated that the model may require refinement.” For one,
the sensitivity decrement predicted by the model does not always occur in
high-event rate/successive-discrimination tasks. Secondly, sensitivity
decrements have been observed in cases (such as high-event-rate/simultaneous-discrimination
tasks) where no such decrement should occur as predicted by the model. Even
more pressingly, studies have shown that a sensitivity decrement can occur even
in the absence of a memory load. This evidence to the contrary prompted See and
other researchers to wonder whether the sensitivity decrement is caused by the total task demand rather than one or
two aspects of the task. In particular, following the study by Koelega et. al., See investigated the
distinction between sensory and cognitive vigilance tasks. In sensory tasks,
the signal is a change in the physical
attributes of the background event, whereas cognitive tasks utilize
symbolic or alphanumeric events and signals. Ultimately, See et. al. found cognitive vigilance tasks
were associated with a “vigilance increment,” a term coined to describe the
stabilization in subjects’ performance as a function of time. This is in direct
contrast to the results of sensory vigilance tasks reported previously. Thus,
on the basis of these results, See et.
al. suggests that the PVT needs to be refined to incorporate the
distinction between cognitive and sensory tasks.
Warm
and colleagues discuss the developments and advances in vigilance research and
argue that evidence from behavioral, subjective, and neural studies support the
conclusion that vigilance requires mental work and is stressful, in accordance
with attentional resource theory. Warm et.
al. cites studies that support the PVT proposed by Parasuraman.
Specifically, he notes that a number of studies have shown that successive
tasks require more mental attentional resources, such as memory, than
simultaneous-discrimination tasks. Indeed, he found that factors known to
increase information-processing demand had a more drastic effect on performance
on successive tasks than on simultaneous tasks, a result that supports the
assertion that successive tasks are more resource demanding. However, Warm and
colleagues also address the criticisms posed against resource theory. Specifically,
the fact that the resources, as well as the increases in resource demands, are
usually inferred from performance
rather than measured (a phenomenon seen in the Parasuraman paper) is a striking
flaw. In response to this criticism, Warm discusses studies that take the
experimental procedures of Parasuraman and see one step further to incorporate
neuroimaging techniques such as PET and fMRI. Studies such as these
demonstrated increased cerebral blood flow in areas of the prefrontal cortex
could serve as proxies to quantify attentional resources and mental workload.
Furthermore, neuroimaging techniques such as transcranial Doppler sonography
(TCD) have shown that the vigilance decrement is associated with a decline in
blood flow velocity. Furthermore, the decline of blood flow was directly
correlated to task demands, especially in the right cerebral hemisphere. The
addition of this neural component provides greater insight into the processes
underlying the vigilance decrement.
QUESTIONS AND COMMENTS
Reading these three papers left me
with several questions and comments on their approach, results, and inferences.
Here, we will first consider the strengths and weaknesses of the Parasuraman
Vigilance Taxonomy, as described by Parasuraman in his 1979 paper, and then
address some further questions on this paper. We conclude with some comments on
the results and inferences of See et. al.
and Warm et. al.
Strengths and Weaknesses
The Parasuraman
Vigilance Taxonomy has several prominent strengths, but also has several
important flaws. Here, we examine the key strengths and weaknesses of this
taxonomy.
Strengths: The
greatest strength of the Parasuraman Vigilance Taxonomy, in my opinion, is the
fact that it provides a basis upon which the vigilance decrement could be
understood and characterized. While no platform is perfect, the PVT gives
researchers something concrete to worth with and expand upon. Furthermore, it
takes into account that different types of vigilance tasks impose different
demands on the subject, and thus may lead to a vigilance decrement for
different reasons. The insight of PVT—that the same observed vigilance
decrement may actually be due to different reasons—is a difficult and rare feat
and serves as one of its key strengths. Furthermore, the fact that PVT is
mostly supported by a variety of vigilance tasks—from auditory to
visual—provides an additional pillar of strength of this taxonomy. Last week, I
questioned why the vigilance tasks utilized were confined to visual tasks and
whether other types of vigilance tasks (such as auditory tasks) would produce
different results. The congruence in the results of the auditory and visual
tasks, and in particular, the fact that PVT takes into consideration different
types of vigilance tasks, is an important strength.
Weaknesses: While
the PVT has several key strengths, it also has important weaknesses. The most
striking flaw for me was its disagreement with some empirical data, as reported
by Parasuraman himself in later papers. For instance, Parasuraman found no
sensitivity decrement in 5 of the 19 high-event-rate/successive-discrimination
tasks; since these tasks clearly load memory, this result is in direct contrast
to the Parasuraman Vigilance Taxonomy. Furthermore, the presence of a
sensitivity decrement in vigilance tasks where no such decrement is predicted
by the model (simultaneous-discrimination/high-event-rate tasks) also violates
the taxonomy. The clear presence of empirical data that disagree with the
taxonomy is concerning; even more pressing, however, is the possibility that
these data points were not given the adequate consideration during the
formation of the Parasuraman Vigilance Taxonomy. If PVT was formed on the
grounds of biased data (where the limited number of data points that did not
agree with the taxonomy were thrown out or given less attention), then this
would serve as the greatest flaw of the taxonomy. Another weakness of the
taxonomy is its binary nature. Specifically, it considers only two factors that
may potentially affect the vigilance decrement; event rate and type of
discrimination. Although this consideration is already a milestone in the study
of vigilance, it does not take into account other factors of the study that may
affect the vigilance decrement. Addition parameters such as overall task load,
as suggested by See et. al., could
contribute to the vigilance decrement. Furthermore, the original PVT does not take into consideration possible cases where both
sensitivity decrement and response criterion are involved in the vigilance
decrement. Although most of the prior research has suggested that these two
factors may operate independently, it is still worthwhile to consider the interplay between these two factors
as well as other parameters in contributing to the vigilance decrement.
Further Comments on Parasuraman, 1979
The Parasuraman paper is clearly a
groundbreaking achievement in the study of vigilance, and I found the paper
extremely interesting. I did have some questions, however, about the inferences
made from the results. For one, I found the attribution of the sensitivity
decrement to memory load rather abrupt. While no one contests the fact that
higher event rates during a successive-discrimination task would load memory
more than a low-event one, there can also be alternative contributions to the
sensitivity decrement. Yet, not such alternative contributions are even
explored. Furthermore, no definitive experiment was done to prove the
assumption that memory-load was the key difference between high-event-rate/successive-discrimination
tasks and other tasks. Combined with the empirical data that disagreed with the
model, this challenged the PVT for me. I would have liked to see more
experiments establishing the link between event-rate and memory load.
See et.
al. and Warm et. al.
The researchers in these two studies
addressed some of the inadequacies of the original PVT and made important
progress. For instance, See et. al.
addressed some of the inconsistencies of the PVT with experimental data and echoed
the question whether there exist other contributions to the sensitivity
decrement. While See et. al. makes an
important revision to the original PVT by pointing out the difference between
cognitive and sensory tasks, it also falls prey to the semi-binary inference-making
seen in the Parasuraman paper. While their results illustrate a clear
difference between cognitive and sensory vigilance tasks, they fail to take
into account the interplay between these two factors. What about tasks that
incorporate both cognitive and sensory elements? Do these two parameters
operate independently or do they interact to produce the sensitivity decrement?
I found the Warm paper extremely
fascinating. For one, it is one of the first papers so far on the study of
vigilance that incorporates a neural component and thus provides more
definitive to support their results. Most compellingly, it discusses techniques
that allow researchers to truly quantify the factors involved in the vigilance
decrement. With neuroimaging and similar techniques, researchers studying
vigilance can stop making inferences based on performance and actually measure factors of interest in this
decrement. However, as painful as it is to say, even neuroimaging and
neuroscience techniques have its limits. It will be interesting to observe and
contribute to the advances in neuro-technology that may ultimately transcend
the present limitations.
LINKS TO NEUROSCIENCE AND FUTURE DIRECTIONS
I think Warm and colleagues have the
correct idea in incorporating evidence and results from a variety of studies,
including neural, behavioral, and subjective. I would be interested to further
explore the neurobiological underpinnings of the vigilance decrement in the
future, with a particular emphasis on the differences in the neurobiological
process between successive-discrimination and simultaneous-discrimination
tasks. Specifically, I think it would be interesting to have a “mouse model” of
this phenomenon. A possible (and rudimentary) experimental design could be as
follows. Since LTP, or long-term potentiation, is the cellular correlate of
long-term memory formation, to investigate the role that memory load plays in
the sensitivity decrement, we can impair early stage LTP or the working memory
of mice and then measure their performance on successive-discrimination and
simultaneous-discrimination tasks. If memory load is indeed the determining
factor, then the mice with impaired LTP should demonstrate a greater
performance decrement in the successive-discrimination trails than the
simultaneous trails. Clearly, there are difficulties that must be overcome to
perform an experiment like this. Mice need to be trained to perform the
difference vigilance trails, and most importantly, humane treatment of the mice
MUST be ensured.
Disclaimer: Okay,
from working with mice, I know lab mice don’t actually look like that, but who
can resist sleeping mice with a teddy bear or a mouse on a swing?
REFERENCES
See,
J.E., Howe, S.R., Warm, J.S., & Dember, W.N. (1995). Meta-Analysis of the
Sensitivity Decrement in Vigilance. Psychological
Bulletin, 117, 230-249.
Warm,
J.S. and Parasuraman, R., et. al. (2008). Vigilance
Requires Hard Mental Work and Is Stressful. Human Factors. 50(3): 433-441.
Image References
Interview
Prep: What is your Greatest Weakness? http://www.medpreps.com/medpreps/interview-prep-greatest-weakness/
Accessed 06/02/2013.
Strength.
Looking for Tigger. http://www.lookingfortigger.com/2012/04/05/strength/
Accessed 06/02/2013.
Ask
a Question About JHD. http://en.hdyo.org/eve/questions/
Accessed 06/02/2013.
Tumblr.
http://www.tumblr.com/tagged/baby%20mouse?before=39
Accessed 06/02/2013
No comments:
Post a Comment