Chapter 2 Literature Review
2.1 Workload and Performance
2.1.4 Relationships between Workload and Performance
This section provides a review on previous models describes relationships between workload and performance, and research investigating effects of higher workload on driving performance. The high workload imposed on drivers was separated into two categories according to their source: those caused by primary driving demand (e.g. complex road conditions or busy traffic), and those due to secondary task demands (e.g. turning on the radio or in-vehicle conversation). When reviewing the effects of secondary tasks, some essential concepts and theories, for example, divided attention and task switching, time sharing and task paralleling are also described.
2.1.4.1 Models of the Relationship between Workload and Performance
A simplified relationship between workload and performance was described by Meister
(Meister, 1976), in which three regions were defined, region A, B and C, see Figure 2.3. Region A is described as when workload is in a relatively low level, even when it increases,
performance can still stay unaffected; Region B represents the area when workload increases upon a certain threshold, performance started steadily decreases with the increasing workload; While in Region C, when workload exceed an upper limit, performance remains at this minimum level, and does not deteriorate even further.
Workload
Performance
Low High Region: A B C Low HighFigure 2.3 The hypothetical relationship between demand and performance
This model only describes the effect of task demand which is “too high” for a human operator. However, other research also suggested that not only highly demanding tasks have negative impact on task performance, but also the low demanding ones, for example, boredom can reduce the capacity by lower operators’ arousal level, and in turn cause a higher mental
workload (Meijman & O’Hanlon, 1984). Later, Brookhuis’s (Brookhuis et al., 1999) studies of drivers’ behaviour under underload and overlaod conditions also suggested that both of these conditions were related to driver performance impairment, as the underload condition causes reduced alertness and lowered attention. De waard (de Waard, 1996a) therefore included a Regions D and A1 into the previous model to form a “Inverted U” Shape, as suggested by Teigen (Teigen, 1994), shown in Figure 2.4. In driving, for example, the under-load condition (Region D) can occur when driving in a dull environment (motorway driving), or when assisted by an adaptive cruise control system, which could reduce a driver’s arousal level and cause fatigue; while the overload condition (Region B and C) can occur when dual-tasking, which can also lead to diverted attention from the road ahead, or insufficient time for critical information processing. Between underload and overload (Region A1, A2 and A3), drivers can compensate for the increased workload by re-allocating effort. Therefore in these Regions, adding further task demand may increase workload without necessarily affecting driver’s performance. When the task demand carries on increasing and exceeds a certain level, a driver will be unable to maintain the performance even with the maximum effort, and driving performance will deteriorate unavoidably, as shown in region B in Figure 2.4.
Another model described the relationship between workload and performance -- Malleable
Attentional Resource Theory was proposed by Young and Stanton to describe the mental
underload when using automated systems (Young & Stanton, 2004). Usually, automated systems perform well when a human operator is able to cope with them. However, they are often ill-prepared to cope with a sudden demand. The poor performance with automation includes the fatigue, poor situational awareness, too much trust, and vigilance decrements. The
Malleable Attentional Resource Theory contains three basic rules: 1) the attentional
resources are malleable; 2) the attentional resources are yoked to task demand; and 3) there is a lag in the attentional resource expansion. In other words, when task demand is reduced (e.g.
when assisted by an automated system), the attentional resource pool shrinks to accommodate the reduced demand (i.e. as being malleable), which is considered to be cognitively efficient by Young and Stanton; and when an increase in demand suddenly occurs to the human operator, e.g. actions required in a semi-automated system, the individual will be unable to cope with this requirement, since the resource pool cannot expand quickly enough (Young & Stanton, 2004). This theory provides an expanded explanation for the effect of underload condition in Region D and A1 in the “inverted U”.
2.1.4.2 Effect of High Workload on Performance
As discussed above, it should be noted that the workload does not always cause performance deterioration. When workload increases, driving performance may decrease, but not always, for example, as suggested in the previous section in Region A. Reid and Nygren also suggested (Reid & Nygren, 1988) that even when an operator’s workload exceeds a certain value, it is the
probability of performance breakdown increase, rather than that there is a value that will definitely result in performance breakdown. To avoid overload, drivers may apply different
strategies, like making more effort, or simplifying the task by limiting the number of information sources they use for decision making (Cooper & Zheng, 2002; Wright, 1974). However, these strategies can also result in impaired driving if a wrong decision is made due to the reduction of information source, for example, neglecting the factor of road surface condition (dry or wet) when distracted (Cooper & Zheng, 2002). High workload could lead to risky behaviour, for example, Hoyos (Hoyos, 1988) also noted that high workload can decrease drivers’ hazard detection, and increase the probability of a driver making risky decisions due to time pressure. This along with other evidences, suggest that high workload may be related directly to high accident risk, for example, Wood et al. (Wood et al., 2003) found in a simulator study, that drivers took longer to brake for critical events (e.g. stop signs or pedestrians at intersections) when they were given the task of counting the number of a certain type of pedestrian. In addition, even when sometimes extra workload appears to be not high enough to
cause driver’s overload or performance deterioration, the cumulative effect of sustaining attention or making effort in a higher level during certain period of time can also be a source of stress which can reduce drivers’ attentional capacity (Hancock & Warm, 1989), which in turn may lead to safety concern.
2.1.4.2.1 Impact of Primary Task Demands on Performance
Workload caused by complex driving conditions, i.e. primary driving related workload, can impact on both secondary and primary task performance. For example, Brown & Poulton (Brown & Poulton, 1961a) reported that subjects’ performance in calculation tasks decreased, in terms of completed fewer, in more complex driving environments than in simpler ones. Lee and Triggs (Lee & Triggs, 1976) reported that drivers missed more target lights in a peripheral detection task while driving through busier and more complex environments. The former also found a lower speed in the complex driving environments (i.e. city traffic) than in simpler ones (residential areas). Harms (Harms, 1991) also found that in more complex driving environments (e.g. different village areas and close to rural junctions), reaction time to calculation tasks was longer, in combination with a lower driving speed than in the control condition (or baseline, i.e. normal driving without performing secondary task), on highway sections. Martens and van Winsum (Martens & Van Winsum, 2000) found negative effects on both reaction time and hit rate for a peripheral detection task when the driving task difficulty was increased by external causes such as narrow curves or the appearance of an unexpected obstacle. Other research on distractions (Engström et al., 2005; Horberry et al., 2006; Jahn et al., 2005; Jamson & Merat, 2005; Tornros & Bolling, 2006) reported that increasing task demand in relatively complex situational environments will impair driving performance even further. For example, in Jamson and Merat’s (Jamson & Merat, 2005) research on distraction due to surrogate IVISs tasks which was conducted in a simulator, both secondary task and primary performance deteriorated, but even more so in more complex road sections (curve or with hazardous events). Especially when performing visual tasks the deviation in lane position only increased by about 2cm compared to baseline (from 20cm to 22cm, values estimated from figure) on straight road sections; however, on curved road sections the measurement increased by 9cm, from 36cm to 45cm.
However, drivers can adapt their behaviour for the workload caused by complex driving conditions, as Senders et al. stated (Senders et al., 1967a), the attentional demand from road environment depends not only on the road condition itself (e.g. layout, geometry, etc), but also on the traffic situation which includes driver’s speed choice and other factors. Therefore the speed limit at which the drivers travelled can be determined by the drivers’ information processing capacity, according to the information generation rate of the road. Drivers tended to
drive at the speed which can balance these two factors (Senders et al., 1967a), for example, drivers behave differently when travelling on familiar or unfamiliar routes.
2.1.4.2.2 Impact of Secondary Task Demands on Driving Performance
The effect of workload induced by the secondary task on driving performance was found in many studies. Either consciously or subconsciously, drivers developed a strategy to reduce the primary task load whilst performing the concurrent secondary task. Harbluk, Noy, Trbovich, Eizenman (Harbluk et al., 2007a) investigated the influence of the mental tasks by arithmetic tasks, and results showed that more occurrences of hard braking were found during the most difficult mental tasks, along with the effect of narrowing of attentional focus. Effects of secondary tasks on driving performance were also reported in research related to mobile phone use (Alm & Nilsson, 1994; Brookhuis et al., 1991; Burns et al., 2002; Poysti et al., 2005; Strayer et al., 2006; Summala et al., 1996; Tornros & Bolling, 2005), In-vehicle entertaining systems (Summala, Nieminen & Punto, 1996), Surrogate IVISs systems, both auditory and visual tasks (Anttila & Luoma, 2005; Engström, Johansson & Östlund, 2005; Jamson & Merat, 2005; Östlund et al., 2004), and other in-vehicle systems (Burnett & Joyner, 1997; Horberry et al., 2006; Jahn et al., 2005; Serafin et al., 2007; Walker et al., 1991; Wierwille & Tijerina, 1996; Wittmann et al., 2006; Zheng et al., 2008).
Driving a vehicle in real traffic is predominately visually demanding and, according to Wickens (Wickens, 1984a) resources are multiple, and therefore simultaneous processing of visual
information for a secondary task causes a structural interference with driving. On the other hand, when tasks are presented verbally and require only vocal responses, they cause less intrusive stimuli and reduce the interference with primary tasks, which should maintain the structural interference to a minimum. However, as mental tasks still conflict with driving in mental resources, drivers may therefore attempt to free up resources by simplifying the primary task, which can be manifested by changes in driving behaviour. Performing visual tasks generally causes more behaviour changes and performance deterioration, in terms of lower speed, higher speed deviation and reversal rate, impaired lane control, delayed braking and less awareness of surroundings, so that drivers react slower or miss potential hazards.
To understand the underlying process and effects of dual-tasking, some essential concepts including attention dividing, task switching, time sharing and task paralleling, are briefly described as following:
Divided attention and task switching
Higher workload caused by secondary tasks is not only due to the task demand itself, but also to the effect of attention being divided between the IVIS systems and the primary task. Divided attention requires the ability to process and/or respond to information while conducting two tasks simultaneously, which is related to a decrease of driving capacity (Lengenfelder et al., 2002). When two tasks are performed simultaneously, the combined workload can be more than the sum of workload of the two tasks, the effect of which is more pronounced for elderly people (de Waard, 1996a). Anderson and colleagues (Anderson et al., 1998) proved the cost of divided attention by interrupting a memory task in two stages (encoding and retrieving), which
suggested that dividing attention at the encoding stage disrupted memory performance, and the reaction time was prolonged in the retrieving stage. The longer reaction time again was more noticeable for older drivers. Schvaneveldt (Schvaneveldt, 1969) showed that even performance on relatively simple tasks can be degraded when they are performed simultaneously with other complex and independent tasks. Switches between tasks were also associated with persistent costs, even when the two tasks are predictable and simple (Rogers & Monsell, 1995). Given that driving already requires attention being divided, including continuously monitoring on-road information, while scanning the environment for potential hazards, and shifting attention onto dashboard as needed (Lengenfelder et al., 2002), adding another task will cause further cost for drivers. In addition, it was found that one the most pronounced effect on multi-tasking is time pressure (Moray et al., 1991), which is an essential element for road safety.
Time sharing
Edquist stated (Edquist, 2008) that to successfully perform the primary driving task, drivers
must be able to select the necessary information and process it in time to make the appropriate decision and execute the required action. Therefore anything that interferes with this process
and/or the speed of this processing, for example performing secondary tasks, will impair the driving performance. The degree of interference between concurrent tasks can be manipulated by the task structure and task priority. As described in Section 2.1.1.2, in the three-dimensional space composed of the information modality (e.g. visual/auditory), the stages of processing (encoding, central processing, responding), and the response types (e.g. manual/vocal), the closer the two tasks are, the more they will interfere to each other due to more competition for shared resources.
Tsimhoni and Green (Tsimhoni & Green, 2003) explained that while performing visual
concurrent tasks, there are four constructs that may affect the time-sharing while driving: 1) the time pressure of looking at the road, 2) the interference of concurrent driving in cognitive
demands, 3) the drivers’ ability of postponing information processing, and 4) the cost of “chunking” the task into smaller parts. Obviously when drivers have to look at something
inside the vehicle, they cannot simultaneously look outside the vehicle, and hence their ability to respond to external events suffers. In addition to the effect of visual-manual conflict, many driving simulator studies have also shown that even non-visual tasks can affect the visual perception as well, because of the mental resource competition.
Task paralleling
Previous research discovered that there is one exception of human’s limitation in information processing in task sharing, which is tasks which can be processed in parallel due to atomisation developed by practice (Shiffrin & Schneider, 1977). The dynamics of visual attention also allows people to shift attention quickly within fixations, including objects in periphery vision. Schumacher suggested that (Schumacher, 2001) skilled procedural decision making and
response selection for two or more tasks can proceed at the same time under adaptive executive control. After some training, some subjects could achieve time sharing in the dual-tasking, and
also, interference between tasks could be adjusted by task priorities and personal preferences. According to Michon’s control-manoeuvre-strategy model (Michon, 1985), the control level as the lowest level of driving behaviour is the highly likely to be performed automatically by experienced drivers. It is possible that experienced drivers can use automated vehicle control when coping with in-vehicle tasks based only on the peripheral view (Summala, Nieminen & Punto, 1996), so that the multi-task demands in normal driving conditions still remain in their attentional capacity.
However, there are limitations of these parallel controls. When driving demands are increased by, for example, traffic density increasing, driving in the intersection, traffic circle, etc., the tasks demands may sometimes exceed driver capabilities (Hancock & Parasuraman, 1992), and the useful visual field still gets narrower with increased mental workload (Jahn et al., 2005).