13	Response Selection and Principles of Compatibility

It’s the le.... It’s the right one

David McClelland

copilot of a British Midland flight that
crashed in January, 1989, before the crew turned off the wrong engine.

INTRODUCTION

Human–machine interaction requires that the operator perceives information, cognitively processes that information, and ultimately selects and executes an action. Even if perception and cognition are accomplished flawlessly, an operator may still take an inappropriate or inaccurate action. The quotation with which this chapter begins illustrates an occasion on which a flight crew turned off the wrong engine, resulting in the crash of a commercial aircraft. This kind of error is called a response-selection error. Response-selection errors cannot be entirely avoided, but proper design can increase the speed and accuracy with which operators can select responses.

In the laboratory, response selection is studied using reaction-time tasks. As we described earlier, such tasks require an observer to be instructed to make a rapid response to a stimulus. We can distinguish between three basic processes that intervene between the onset of a stimulus and the completion of the response initiated by the stimulus (see Figure 13.1): stimulus identification, response selection, and response execution. How well stimulus identification can be performed is a function of stimulus properties such as brightness, contrast, and so forth. How well response execution can be performed is a function of response properties such as the complexity and accuracy of movements that must be made. The focus of this chapter, response selection, is on how quickly and accurately people can determine which response they are to make to a stimulus. How well response selection is performed is influenced primarily by the relationships between the members of the stimulus set and the responses assigned to each.

FIGURE 13.1The three-stage model for reaction-time tasks.

If a person is expected to operate a machine effectively, the interface through which he or she controls the machine must be designed to optimize the efficiency with which displayed information can be transformed into the required controlling responses. Understanding the processes involved in response selection is critical to our understanding of action more generally. This chapter will discuss those factors that influence the time taken to choose between different responses and how experiments are designed to evaluate alternative interface designs.

SIMPLE REACTIONS

Some tasks require an operator to react to a signal with a single, predetermined response (Teichner, 1962). For example, if an alarm sounds, the operator may have to press a button to shut off a piece of equipment as quickly as possible. Situations in which a single response must be made to a stimulus event are called simple reaction tasks. Response-selection processes are presumed to play a minimal role in simple reaction tasks, because only one response is to be made to any event: there are no choices among responses to be made (Miller, Beutinger, & Ulrich, 2009). However, even for a simple reaction, a decision still must be made about the presence or absence of the stimulus itself (Rizzolatti, Bertoloni, & Buchtel, 1979).

It helps to understand what is happening in a simple reaction task by considering a model of the response process in which evidence about the presence of the stimulus accumulates over time and is stored somewhere in the brain. The observer can execute a response as soon as he or she has obtained enough evidence that a stimulus is present (e.g., Diederich & Busemeyer, 2003; Miller and Schwarz, 2006). In some situations, the observer will need a lot of evidence before deciding to respond. For example, if the observer’s task is to shut down a large machine and interrupt a production line, she may need to feel very sure that the signal has occurred before responding. In other situations, the observer may not need as much evidence. So, even though there is no response selection to be made in a simple reaction task, the reaction time will still be influenced by the amount of evidence that the response requires.

Reaction time in a simple reaction task will be more strongly affected by stimulus factors. As the stimulus is made more salient, reaction time will decrease. There are several ways that a stimulus can be made more salient. For example, as the intensity, size, or duration of a visual auditory or tactile stimulus is increased, reaction time to its onset will decrease (Miller & Ulrich, 2003; Schlittenlacher & Ellermeier, 2015; Teichner & Krebs, 1972). The fastest that a simple reaction can be is approximately 150 ms for visual stimuli and 120 ms for auditory and tactual stimuli (Boff & Lincoln, 1988). Referring again to a model in which evidence accumulates over time, stimulus factors such as intensity affect primarily the rate at which information about the presence of the stimulus accumulates. Consistently with such a model, reaction time is shorter when two or more redundant signals (e.g., an auditory and a visual stimulus) are presented simultaneously (Miller et al., 2009), allowing a more rapid accumulation of information.

Another major factor influencing simple reaction time is whether or not the observer is prepared to respond to a stimulus. When an observer is unprepared, his response times will be longer than when he is prepared. In some situations (as in most laboratory experiments), an observer can be prepared for, or warned about, an imminent signal. A warning signal can increase an operator’s state of readiness, lowering the amount of evidence necessary for responding. However, preparation to respond is attention-demanding, and when an observer must perform another task simultaneously, simple reactions will be slowed significantly (Henderson and Dittich, 1998).

A task that is closely related to the simple reaction task is a “go/no go” task (see our discussion of Donders in Chapter 1). In this task, a person responds only when a “target” stimulus occurs but does not respond to any other stimulus. The person can make two kinds of errors in performing this task: omissions (failing to respond to a target stimulus) or commissions (responding to the other stimulus; see Chapter 3). Omission errors are sometimes attributed to failures of attention, whereas commission errors are thought to be due to failures of response inhibition (Bezdjian, Baker, Lozano, & Raine, 2009).

CHOICE REACTIONS

Usually, an operator will need to choose one of several possible alternatives when an action is required. Situations in which one of several possible responses could be made to a stimulus are called choice reaction tasks. In most choice reaction tasks, many responses could be made to many different stimuli. For example, several distinct auditory alarm signals, each with its own assigned response, may be used in a process control room. When an alarm sounds, the control room operators must identify it and choose the appropriate course of action.

Response selection has been investigated most thoroughly for tasks in which one of two or more responses is to be made to one of two or more possible stimuli. Most of the variables that affect simple reaction time, such as stimulus intensity and state of readiness, also influence choice reaction time. However, other factors are also important, including the relative emphasis on speed or accuracy, warning interval, amount of uncertainty, compatibility of stimuli and responses, and practice.

SPEED–ACCURACY TRADEOFF

In a simple reaction task, an inappropriate response cannot be selected, because there is only one response; errors are made by either not responding or responding at the wrong time. However, in choice reaction tasks, an inappropriate response can be made. The time to make a response in a choice reaction task depends on how accurate the choice must be. For example, if accuracy is of no concern, you could make any response you wanted whenever you detected the onset of a stimulus (extreme speed emphasis in Figure 13.2). You would be simply guessing at the appropriate response, which you could then make very quickly, but your accuracy would be no better than chance. Alternatively, you could wait until you were sure about the identity of the stimulus and its associated response (extreme accuracy emphasis in Figure 13.2). Your responses would be much slower, but your accuracy would be nearly perfect.

FIGURE 13.2A theoretical speed–accuracy tradeoff function.

Between these two extremes, you could choose any level of speed, resulting in some level of accuracy (or vice versa). The function relating speed and accuracy shown in Figure 13.2 is called the speed–accuracy tradeoff (Osman et al., 2000). The speed–accuracy tradeoff function shows different combinations of speed and accuracy that can be obtained for a single choice situation, much as a receiver operating characteristic (ROC) curve shows different hit and false-alarm rates for a given sensitivity (see Chapter 4). As with the ROC curve, the selection of a point on the speed–accuracy function is affected by such things as instructions and payoffs. A person’s speed–accuracy criterion can be manipulated experimentally by imposing different response deadlines or presenting a signal to respond at varying delays after stimulus onset (Vuckovic, Kwantes, Humphreys, & Neal, 2014).

One of the easiest ways to account for the speed–accuracy tradeoff is through information accumulation models similar to the one we presented for simple reaction time tasks (Ulrich, Schröter, Leuthold, & Birngruber, 2015). Assume that evidence accumulates and is stored for each response separately, and a response is selected when enough evidence for it has accumulated. Responses made on the basis of less information will be faster but less accurate. Moreover, just as the criterion in signal-detection theory reflects an observer’s bias to respond “yes” or “no,” the threshold amounts of evidence across the alternative responses reflect an observer’s bias. The lower a threshold is in relation to other thresholds, the greater is the bias toward that response. For human factors, the primary implication of the speed–accuracy tradeoff is that variables may affect either the efficiency of information accumulation or the threshold amount of evidence required for each response.

Alcohol has a marked effect on a person’s speed–accuracy tradeoff (Rundell & Williams, 1979). In a study investigating effects of alcohol, people performed a two-choice task in which they were to make left or right keypresses to low- or high-frequency tones. Asking them to respond at three different speeds (slow, medium, and fast) generated a picture of each person’s speed–accuracy tradeoff function. People who had drunk alcohol prior to performing the task made their responses just as quickly as people who had not, but they had higher rates of error. The shape of the speed–accuracy tradeoff functions suggested that alcohol impaired how well information accumulated. These results suggest that an alcohol-impaired person will have to respond more slowly to avoid making mistakes.

TEMPORAL UNCERTAINTY

As we described for simple reactions, knowledge about when a stimulus is going to occur affects the speed of responding in choice reactions. If a person knows that a stimulus will be occurring at a particular time, he can prepare for it. Warrick, Kibler, and Topmiller (1965) had secretaries respond to the sound of a buzzer by reaching to and pressing a button located to the left of their typewriters. Half the secretaries were given warning that the buzzer was going to sound, and the other half were not given warning. The buzzer was sounded once or twice a week for 6 months. Their reaction times decreased over the 6-month period, but secretaries who were not warned were always approximately 150 ms slower to respond than secretaries who were warned.

The effects of preparation are examined in the laboratory by varying the time between a warning signal and the stimulus to which an observer is to respond. We can plot a “preparation function” for such a task by showing the response times as a function of the time between the warning signal and the stimulus. The data shown in Figure 13.3 come from a study by Posner, Klein, Summers, and Buggie (1973). They asked observers to respond with a left or a right key to an X that occurred to the left or right of a vertical line. On each trial, there was either no warning or a brief warning tone, followed by the X after a variable delay. The left panel of the figure shows that reaction times were a U-shaped function of warning interval, with the fastest responses occurring when the warning tone preceded the stimulus by 200 ms. This function, in which reaction time decreases to a minimum at a warning interval of 200–500 ms and then increases as the interval becomes longer, is fairly typical. Both psychophysiological and behavioral evidence suggests that the U-shaped preparation function is not due to processes involved in executing the response (in the motor stage; Bausenhart, Rolke, Hackley, & Ulrich, 2006; Müller-Gethmann, Ulrich, & Rinkenauer, 2003).

FIGURE 13.3Reaction times (a) and percentages of error (b), as a function of foreperiod, in a two-choice task.

However, note that, as the right panel illustrates, the percentages of errors in this study were the inverse of the reaction time function. Errors increased and then decreased as the warning interval increased. Slower response times were associated with higher accuracy, demonstrating a speed–accuracy tradeoff. In terms of information accumulation models of the type described earlier, this outcome implies that being prepared to respond results in a lower threshold for responding but not in improved efficiency of information accumulation.

Warning signals are sometimes referred to as alerting signals in operational environments (e.g., Gupta, Bisanz, & Singh, 2002). The effectiveness of such signals, or how well they improve responses to an event, will depend on how much time the operator has to respond. If the response must be made very rapidly, processing an alerting signal will use up some of the time that could have been used for processing and responding to the event. Simpson and Williams (1980) had airline pilots respond to a synthesized speech message while flying a simulator. Reaction time measured from the onset of the message was faster when an alerting tone occurred 1 s prior to the message than when one did not. However, when the additional time for the alerting tone was taken into account, overall system response time was actually longer (see Figure 13.4). Consequently, Simpson and Williams concluded that an alerting tone probably should not be used with synthesized speech displays in the cockpit environment.

FIGURE 13.4Mean times to respond to a warning message alone (a) or preceded by an alerting tone (b).

STIMULUS-RESPONSE UNCERTAINTY

Our discussion of stimulus-response uncertainty necessarily involved discussion of information theory, which expresses the amount of information (H) in a set of stimuli or responses as a function of the number of possible alternatives and their probabilities. Information theory is described in more detail in Appendix 2. It became popular in psychology in the 1950s because choice reaction time was shown to be a linear function of the amount of information transmitted, a relation that has come to be known as Hick’s law, or the Hick–Hyman law, after the researchers who first discovered it (see Proctor & Schneider, 2018, for a detailed description of the impact of the law on research and interface design). To understand the Hick–Hyman law, it is important to understand that “information” as it is computed in information theory is a measure of uncertainty: Reaction time increases as uncertainty increases.

Hick (1952) used a display of 10 small lamps arranged in an irregular circle and 10 corresponding response keys on which a person’s fingers were placed. When one of the lamps came on, the corresponding keypress response to that stimulus was to be made. In different sets of trials, the number of possible stimuli (and responses) was varied from 2 to 10. Reaction time was a linear function of the amount of information in the stimulus set (see Figure 13.5).

FIGURE 13.5Reaction times plotted as a function of number of alternatives on a log₂ scale.

In Hick’s (1952) experiment, the stimuli were equally likely, and performance was measured for sets of trials in which no errors were made. If no errors are made, information is perfectly transmitted, and the amount of information transmitted is equal to the stimulus information. If accuracy is not perfect, the amount of information transmitted depends on the stimulus-response frequencies (see Chapter 4). Therefore, in a second experiment, Hick encouraged people to respond faster but less accurately. The decrease in reaction time was consistent with the reduction in information transmitted.

Information (uncertainty) also decreases when some stimuli are made more probable than others. Hyman (1953) showed that when uncertainty was reduced in this way, or by introducing sequential dependencies across trials (which altered the probabilities on a trial-to-trial basis), average reaction times were still a linear function of the amount of information transmitted.

The Hick–Hyman law is written as

$$\text{Reaction time}=a+b\left[T\left(S,R\right)\right]$$,

where:

a	is a constant reflecting sensory and motor factors,
b	is the time to transmit one bit of information, and
T(S,R)	is the information transmitted (see Figure 13.5).

The Hick–Hyman law fits data from many other choice reaction tasks, but the rate of information transmission (1/b) varies greatly as a function of the specific task. The general point of the Hick–Hyman law is that in most situations, an operator’s reaction time will increase by a constant amount for each doubling of the number of distinct possible signals and associated responses. The fewer the alternatives, the faster the operator can respond.

The studies by Hick (1952) and Hyman (1953) led to a resurgence of interest in choice ­reaction time, which had been studied extensively in the latter part of the 19th century and the early part of the 20th century. Much of the research following Hick and Hyman placed qualifications on the Hick–Hyman law. For instance, Leonard (1958) showed that although the Hick–Hyman law describes the effect of the information on reaction time, a model based on binary decisions of the type implied by information theory cannot explain how people select responses in choice-reaction tasks.

Usher, Olami, and McClelland (2002; see also Usher & McClelland, 2001) provided evidence that the Hick–Hyman law is a consequence of subjects attempting to maintain a constant level of accuracy for all stimulus-response set sizes. They assumed that people used the information accumulation mechanism discussed above to select one of N possible responses: the response for which evidence reaches threshold first is selected. Lower thresholds result in faster responses than higher thresholds, because a low threshold is reached sooner after stimulus onset. However, alternatives with lower thresholds will be more easily reached regardless of what kind of information is coming in, so these alternatives will be more likely to be selected by mistake, resulting in a higher error rate.

When the number of response alternatives increases, another accumulator mechanism must be added to those already operating. Each additional alternative therefore adds another opportunity for an incorrect response to be selected. To prevent the error rate from increasing to potentially very high rates with many response alternatives, all the response thresholds must be adjusted upward. Usher et al. (2002) showed that the probability of an error will not increase if the increase in the threshold as N increases is logarithmic. This logarithmic increase in thresholds results in a logarithmic increase in reaction time. The Hick–Hyman law is therefore a byproduct of a person’s attempts not to make too many mistakes as the number of possible responses increases.

Although reaction time typically increases as a function of the number of stimulus-response alternatives, the slope of the Hick–Hyman function is not constant and can be reduced to approximately zero with sufficient practice (Mowbray, 1960; Mowbray & Rhoades, 1959; Seibel, 1963). Moreover, for highly compatible stimulus-response mappings, the slope approximates zero even without practice. Leonard (1959) demonstrated this using stimuli that were vibrations to the tip of a finger and responses that were depressions of the stimulated fingers. With these tactile stimuli, there was no systematic increase in reaction time as the number of choices was increased from two to eight. We see the same lack of effect of stimulus-response uncertainty on reaction time for responses requiring eye movements to the locations of visual stimuli (Kveraga, Boucher, & Hughes, 2002) or aimed movements to a target (Wright, Marino, Chubb, & Rose, 2011), and for responses that require saying the names of the digits or letters (Berryhill, Kveraga, Webb, & Hughes, 2005). Thus, the number of alternatives has little effect for highly compatible or highly practiced stimulus-response relations—topics to which we turn next.

PRINCIPLES OF COMPATIBILITY

STIMULUS-RESPONSE COMPATIBILITY

About the same time as the work of Hick and Hyman, Fitts and Seeger (1953) reported another classic choice reaction time study. They used three different stimulus and response sets, all with eight alternatives. Thus, the sets were equivalent in terms of the amount of stimulus and response information they contained. The stimulus sets differed in the way that the information was signaled (see Figure 13.6). For set A, any one of eight lights could come on; for stimulus sets B and C, any of four lights alone or any four pairs of these lights could occur. The three response sets corresponded conceptually to the displays. People were required to move a single stylus to a target location for sets A and B, and two styli to locations for set C. For response set A, there were eight locations in a circular configuration; for response set B, there were also eight locations, but responses were made by moving from the start point along one of four pathways, which then branched into Ts; for response set C, there were up-down locations for the left hand and left-right locations for the right hand, with the eight responses signaled by combinations of the two hand movements.

FIGURE 13.6Stimulus sets, response sets, and data (reaction time and error rates).

Responses in this study were faster and more accurate for the pairings of stimulus sets and response sets that corresponded naturally than for those that did not. Fitts and Seeger called this phenomenon stimulus-response (S-R) compatibility and attributed it to cognitive representations or codes based on the spatial locations of the stimulus and response sets.

In a second classic study, Fitts and Deininger (1954) manipulated the mapping of stimuli to responses within a single stimulus and response set (the circular sets from the previous study). The operator’s task was to move the stylus to an assigned response location when one of the stimuli was lit. There were three stimulus-response assignments: direct, mirrored, and random. With the direct assignment, each stimulus location was assigned to the corresponding response location. With the mirrored assignment, the left-side stimulus locations were assigned to their right-side counterparts of the response set, and vice versa. Finally, with the random assignment, no systematic relation existed between the stimuli and their assigned responses. Responses were faster and more accurate with the direct assignment than with the mirrored assignment. Even more strikingly, the reaction times and error rates for the random assignment were over twice those for the mirrored assignment.

Morin and Grant (1955) further explored the effects of relative compatibility by having people respond to eight lights arranged in a row with combinations of keypresses. Responses were fastest when there was a direct correspondence between stimulus and response locations (see Figure 13.7). Responding was also fast when the stimulus and response locations were perfect mirror images of each other (i.e. the rightmost response would be made if the leftmost stimulus occurred, and so on), as in Fitts and Deininger’s (1954) study. When the stimulus-response assignments were quantified by a correlation coefficient, they found that reaction time increased as the correlation approached zero. These findings indicate that people can translate quickly between a stimulus and response when a simple rule (e.g., respond at the location opposite to the stimulus location) describes the relation (Duncan, 1977). It has been suggested that the correlation coefficient might provide a metric for compatibility in real-world situations for which several stimuli and responses are involved (Kantowitz, Triggs, & Barnes, 1990).

FIGURE 13.7Reaction time as a function of the correlation between stimulus and response locations.

Since these early experiments on S-R compatibility, many basic and applied studies of compatibility effects have been completed (see Proctor & Vu, 2016). These compatibility studies have implications for the design of displays and control panels. The most important of these implications is the fact that machine operation will be easiest when the assignment of controls to display elements is spatially compatible. A classic applied illustration of the compatibility principle comes from studies of four-burner ranges conducted by Chapanis and Lindenbaum (1959) and Shinar and Acton (1978). A common design is to have two back burners located directly behind two front burners with the controls arranged linearly across the front of the range (see Figure 13.8b-e, panels 2, 3, and 4). For this design, there is no obvious relation between controls and burners, so a cook may be confused about which control operates a particular burner. However, if the burners are staggered in a sequential left-to-right order (see Figure 13.8a, panel 1), each control location corresponds directly to a burner location, and confusion about the relation between controls and burners is eliminated.

FIGURE 13.8Control–burner arrangements of a stove.

FIGURE 13.9Compatible (a and c) and incompatible (b and d) stimulus-response assignments in two-choice tasks, with the hands uncrossed (a and b) and crossed (c and d).

RELATIVE LOCATION CODING

A widely used procedure for studying S-R compatibility is a two-choice task in which visual stimuli are presented to the left or right of a central fixation point (see Figure 13.9a,b). In the compatible condition, the observer is to respond to the left stimulus with a left keypress and to the right stimulus with a right keypress (see Figure 13.9a). In the incompatible condition, the assignment of stimulus locations to response keys is reversed (see Figure 13.9b). Responses in the two-choice task are faster and more accurate when they are compatible with the stimulus locations (Proctor & Vu, 2016).

S-R compatibility effects also occur when stimulus location is irrelevant for determining the correct response (Hommel, 2011; Proctor, 2011). For example, if an operator is to respond to the color of an indicator light, the location of the light will have an effect on the operator’s response. More concretely, suppose you are to make one response to a red light and another to a green light, and that these lights can occur to the left or right of a central point. If you are to respond to the red light by pressing a key on the right and to the green light by pressing a key on the left, your responses will be fastest when the red light occurs to the right or the green light to the left. The influence of the spatial correspondence between the locations of the lights and the responses is called the Simon effect (Simon, 1990). Note that the Simon effect is a special case of S-R compatibility that arises when the location of the stimulus is irrelevant to the response that is to be made. We will discuss the Simon effect again later in this chapter.

When a person’s left and right hands operate the left and right response keys (as shown in panels a and b of Figure 13.9), the distinction between left and right response locations is redundant with the distinction between left and right hands. This means that we do not know whether S-R compatibility is due to the locations of the responses or to the hands used to make those responses. To determine which is more important, the left and right hands can be crossed, so that the left hand is at the right response location and the right hand is at the left location (see Figure 13.9c,d). With this arrangement, responses will still be faster when there is a direct correspondence between the stimulus and response locations (e.g., Brebner, Shephard, & Cairney, 1972; Roswarski & Proctor, 2000; Wallace, 1971). This means that response location is more important than the hand used to execute the response. This benefit of correspondence between stimulus and response positions occurs even when a person’s hands are not crossed but hold sticks that are crossed to press the opposite-side keys (Riggio, Gawryszewski, & Umiltà, 1986).

Spatial S-R compatibility effects will be observed in many situations (Heister, Schroeder-Heister, & Ehrenstein, 1990; Reeve & Proctor, 1984), even when “left” and “right” stimulus or response locations are not defined relative to the locations on a person’s body (Nicoletti, Anzola, Luppino, Rizzolatti, & Umiltà, 1982; Umiltà & Liotti, 1987). In other words, it is not the absolute physical locations of the stimuli and responses that determine the degree of compatibility, but their locations relative to each other. These findings and those above suggest that stimuli and the locations at which an action is effected are mentally coded categorically (e.g., left or right) by their relative positions (Umiltà & Nicoletti, 1990).

S-R compatibility effects also occur when stimuli and/or responses are not located in physical space. For example, a person will respond faster to the words “left” and “right” when they are assigned to left and right keypresses or “left” and “right” vocal responses (Proctor, Wang, & Vu, 2002). Similar results are obtained with left- and right-pointing arrows presented in a fixed location (Miles & Proctor, 2012) and even when the stimuli vary along a vertical dimension (up or down positions) and the responses along a horizontal dimension (left or right keypresses or movements; Cho & Proctor, 2003). Moreover, compatibility effects arise for a variety of nonspatial stimulus and response dimensions, such as numerosity (one or two brief stimulus occurrences mapped to responses for which a single key is tapped once or twice; Miller, Atkins, & Van Ness, 2005). These results imply that S-R compatibility effects occur whenever there is similarity in the cognitive representations of the stimulus and response sets.

THEORETICAL INTERPRETATIONS

Most explanations of S-R compatibility rely on the idea that stimuli and responses are coded in terms of perceptual and conceptual features. S-R compatibility arises to the extent that the feature dimensions are similar between the stimulus and response sets (called dimensional overlap; Kornblum, 1991). Compatibility effects occur whenever S-R dimensions overlap conceptually. For example, both stimuli and responses could be defined by the concepts of left and right, and a compatibility effect would occur even if the stimuli were the words “left” and “right” and the responses were left and right keypresses. However, compatibility effects are strongest when the dimensions are also physically similar. For example, changing the response to the words “left” and “right” from keypresses to the spoken words “left” and “right” would give a larger compatibility effect (Proctor & Wang, 1997).

Kornblum and Lee (1995) described another way in which stimuli and responses can be similar, which they called structural similarity. Structural similarity can be illustrated in the following example. The letters A, B, C, and D and the numbers 1, 2, 3, and 4 have a common natural order that does not depend on conceptual or physical similarity. The letters A, B, C, and D might be used as stimuli and assigned arbitrarily to the spoken responses “1,” “2,” “3,” and “4.” Responses will be fastest and most accurate when the assignment is consistent with the natural order: A to “1,” B to “2,” etc. Structural similarity may be the reason why a variety of compatibility effects arise in tasks for which people have to make binary decisions and the stimulus and response dimensions are unrelated (as, for example, with up and down stimulus locations mapped to right and left responses; see Proctor & Xiong, 2015).

“single-route” models of S-R compatibility focus on the computations that a person makes in selecting a response when stimulus and response dimensions overlap. In these models, response selection is an intentional process of selecting the instructed response. The most well-defined of these models can be used to predict a person’s performance in applied settings.

Rosenbloom (1986; Rosenbloom & Newell, 1987) developed a model of S-R compatibility that attributes compatibility effects to the number of transformations that have to be performed to select the response. Their model “is based on the supposition that people perform reaction-time tasks by executing algorithms (or programs) developed by them for the task” (Rosenbloom, 1986, p. 156). This kind of model is commonly referred to as a GOMS model (e.g., Kieras, 2004), where GOMS stands for Goals, Operators, Methods, and Selection rules (see Chapter 19 and Box 3 in Chapter 3). When this kind of model is used to explain compatibility effects, a researcher must first perform a task analysis to determine the algorithms that can be used to perform the task. Tasks with low S-R compatibility will require the use of algorithms with more steps than will tasks with high compatibility. Following the task analysis, the researcher must estimate the time required for each operation in the algorithm. Based on these estimates, the researcher can predict how large the compatibility effects on response time will be. Box 13.1 describes an application of this approach to predicting the extent of compatibility for alternative mappings of abbreviations to command names, a problem in human–computer interaction.

Although single-route models can account for many compatibility effects, they do not provide an account of phenomena like the Simon effect, for which a stimulus dimension is defined as irrelevant for the task. Consequently, the most successful models of S-R compatibility are “dual-route” models. These models include not only an intentional process but also an automatic response-selection mechanism.

Kornblum, Hasbroucq, and Osman (1990) proposed a dual-route dimensional overlap model in which a stimulus automatically activates the most compatible response, regardless of whether or not that response is the correct one. The correct response is identified by way of the intentional response-selection route. If the automatically activated response is not the same as the one identified by the intentional route, it must be inhibited before the correct response can be programmed and executed. Inhibition of the automatically activated response when it conflicts with the correct response explains the Simon effect. Compatibility effects for relevant stimulus dimensions arise in the model from both response inhibition and the time required by the intentional response-selection route.

Hommel, Müsseler, Aschersleben, & Prinz (2001) developed the Theory of Event Coding to explain perception–action relationships more generally, including S-R compatibility effects. This theory assumes that codes for stimuli and responses share the same cognitive system that subserves perception, attention, and action. The theory emphasizes structures called event codes or event files (Hommel, 2004; Hommel et al., 2001). A file is a temporary, linked collection of features that define an event (Kahneman, Treisman, & Gibbs, 1992). Hommel et al. proposed that a stimulus and its associated response are coded as linked and integrated features in an event file. The features that are bound into the event file will be less available for other perceptions and actions.

The theory of event coding emphasizes that actions are coded in terms of their effects. Kunde (2001) performed an experiment in which subjects responded to one of four colored stimuli with a keypress made with two fingers on each hand. Pressing a key lit up one box in a row of four on the lower part of the display. The box that lit up is called the effect of the response. In one condition the location of the box corresponded to that of the key, whereas in another it did not. Responses were faster with the corresponding response–effect mapping than with the noncorresponding mapping, even though the effect did not occur until after the key was pressed. If no effects were coded for the actions, the location of the box should not have influenced response times; the fact that it did provides strong support for the theory. We will see more evidence for the role of effects in the control–display population stereotypes we discuss later, in which people anticipate a particular outcome when they operate a control.

S-C-R COMPATIBILITY

Most research on S-R compatibility has focused on situations in which simple rules relate stimuli to responses (Kantowitz et al., 1990). However, by attributing compatibility effects to the cognitive codes used to represent the stimulus and response sets, the implication is that central cognitive processes must be responsible for the effects. Because the role of cognitive mediation in response selection will be larger for more complex tasks that do not involve simple rules or response tendencies, Wickens, Sandry, and Vidulich (1983) have used the term S-C-R compatibility to emphasize the central processes. The mediating processes (C) reflect the operator’s mental model of the task. Compatibility will be observed to the extent that stimuli and responses correspond with the features of the mental model.

Wickens et al. (1983) structured their theory of S-C-R compatibility around the multiple-resources view of attention and, hence, stressed the importance of the cognitive codes (verbal and visual) used to represent the task. They proposed that codes must be matched with input and output modes for S-C-R compatibility to be maximized. Wickens et al. provided evidence that tasks represented by a verbal code are most compatible with speech stimuli and responses, whereas tasks represented by a spatial code are most compatible with visual stimuli and manual responses. Robinson and Eberts (1987) obtained evidence consistent with the coding relations proposed by S-C-R compatibility in a simulated cockpit environment. Either a synthesized speech display or a picture display was used to communicate emergency information. As predicted by the S-C-R compatibility hypothesis, when responses were made manually, they were faster for the picture display than for the speech display.

Greenwald (1970) proposed that the highest compatibility between stimuli and responses occurs when they have high ideomotor compatibility. Ideomotor feedback refers to the sensations resulting from an action. Stimuli and responses have high ideomotor compatibility when the modality of the stimulus is the same as the ideomotor feedback from the response. Ideomotor compatibility is high, for example, when spoken responses (e.g., saying “A”) are required to auditory letters (e.g., “A” played over headphones). Greenwald conducted an experiment in which he presented letters either auditorily and visually and asked people to respond with the name of the letter either vocally or in writing. He found that response times were fastest when auditory letters were paired with naming responses or visual letters were paired with written responses. When the assignment of response modality to stimulus modality was reversed, response times were slower.

The concept of S-C-R compatibility was expanded by Eberts and Posey (1990) to emphasize the role that mental models play in explaining compatibility effects. Specifically, they propose that a good mental model that accurately represents the conceptual relations of the task should enable better and more efficient performance than a poor mental model. Eberts and Schneider (1985) performed several experiments with a difficult control task and showed that people perform better when their training incorporates an appropriate mental model. From experiments such as these and John and Newell’s (1990) work on compatibility effects in human–computer interaction, described in Box 13.1, we know that compatibility effects influence performance in a broad range of tasks that are more complex than those performed in the laboratory.

PRACTICE AND RESPONSE SELECTION

Like any other human activity, performance on choice reaction tasks improves with practice. However, as noted in Chapter 12, benefits will be small when the mapping of stimuli to responses varies. For example, the late psychologist Richard Gregory noted,

There’s the famous case about a mine trolley that had a brake pedal and accelerator—so the accelerator was on the right and the brake pedal on the left for one direction of travel—but reversed on the way back—as one sat on a different seat and used the same pedals. Believe it or not, they had an awful lot of crashes. (p. 38, quoted in Reynolds & Tansey, 2003)

The improvement in performance that occurs when the S-R mapping does not vary on a regular basis is characterized by the power law described in Chapter 12 (also see Newell & Rosenbloom, 1981). This means that although performance continues to improve indefinitely, the additional benefit for a constant amount of practice becomes less and less as people continue to perform the task. Moreover, because practice effects are larger for tasks that have more stimulus-response alternatives, the slope of the Hick–Hyman law function relating reaction time to number of alternatives becomes progressively smaller as people become more practiced (Teichner & Krebs, 1974).

A classic study that illustrates the extent to which human performance can improve with practice was conducted by Crossman (1959). He examined the time required for people to make cigars on a hand-operated machine. Operators were tested whose experience levels ranged from novice to 6 years on the job. Speed of performance increased as a power function of experience up to 4 years, at which point the operators worked faster than the machine, and so the machine processing time determined performance time. In other words, the machine reached its limits before the operators reached theirs.

Seibel (1963) performed a 1023-choice reaction task in which a subset of 10 horizontal lights were illuminated. In response to each presentation of lights, Seibel (who was his own subject) pressed each of the response keys indicated by the lights as a chord. Initially, his reaction times averaged over 1 s, but they decreased to 450 ms after 70,000 trials of practice. As can be seen in Figure 13.10, performance improved continuously over the course of the study.

FIGURE 13.10Reaction time as a function of practice in a 1023-choice task.

There are several explanations of how response-selection processes change as a person becomes skilled at a task. The power law seems to show a gradual, continuous change from unpracticed to practiced states, leading some people to treat practice effects as quantitative changes over time. For example, using the idea of production systems that we discussed in the last chapter, the change might arise from increasing the number of procedures that are incorporated into a single chunk (e.g., Rosenbloom, 1986). Alternatively, a qualitative change over time may occur, such that the procedures used to do the task at the end of practice are not the same as those that were used at the beginning of practice (e.g., Teichner & Krebs, 1974).

Not only do responses get faster, but S-R compatibility effects also decrease with practice. However, they never completely go away (Dutta & Proctor, 1992). For example, Fitts and Seeger (1953) found that responses for an incompatible display–control arrangement were still considerably slower than those for a compatible arrangement after 26 sessions of 16 trials for each arrangement. Such findings are consistent with proposals by Eberts and Posey (1990) and Gopher, Karis, and Koenig (1985) that mental representations continue to play an important role in translating between stimuli and responses even for well-practiced performers. Although we still do not know why S-R compatibility effects persist, the point to remember is that incompatible display–control arrangement can result in decrements in performance that cannot be remedied entirely by practice.

IRRELEVANT STIMULI

The previous section discussed compatibility effects on performance due to features of the stimulus and response sets that were relevant to the task. Earlier, we also discussed the Simon effect, which is similar to compatibility effects except that the factors resulting in a Simon effect are irrelevant to the task.

Simon’s (1990) original experiments used auditory stimuli: People made a left or right response as determined by the high or low pitch of a tone that occurred in the left or right ear. Simon initially proposed that the effect reflects an innate tendency to orient, or respond, toward the auditory stimulus (Simon, 1969). When this response tendency conflicts with the correct response, for example, when a stimulus detected on the left requires a response to the right, the tendency to respond to the left must be inhibited before the correct response can be made. Other explanations of the Simon effect maintain this response competition idea and link it to automatic activation of the spatial response code corresponding to the stimulus location, which produces interference when the wrong codes are activated (Umiltà & Nicoletti, 1990; see our earlier discussion of Kornblum et al.’s, 1990, model).

Why is stimulus location processed even though it is defined as irrelevant to the task? The answer seems to be that the act of discriminating left and right response alternatives activates a similar coding of stimulus locations (Ansorge & Wühr, 2004), causing stimulus location to be added into the information used during the decision process (Yamaguchi & Proctor, 2012). The additional information tends to activate the corresponding response location code. These kinds of activations between codes for stimuli and responses have been attributed to highly overlearned, and possibly even innate, associations between the stimulus and response locations (e.g., Barber & O’Leary, 1997).

Like other S-R compatibility effects, the Simon effect persists even after considerable amounts of practice. However, the Simon effect can be reversed by giving people prior practice in performing a two-choice task in which the response locations are mapped incompatibly to the stimulus locations (Proctor & Lu, 1999; Tagliabue, Zorzi, Umiltà, & Bassignani, 2000). Thus, an operator’s experience in specific contexts may act to reverse the normal advantage for spatial correspondence that a designer might expect to find.

An important consideration is how location of stimuli and responses is represented or coded relative to a person’s body or relative to the machine she is operating. For instance, a stimulus location can be coded as left or right of person’s body midline, or above or below a line on a computer screen. Factors like the relative salience of different reference frames and instructions emphasizing one frame over another can determine which reference frame will affect performance the most.

For example, while driving, turning the steering wheel clockwise will usually result in the car turning to the right. However, when the wheel is held at the bottom, a clockwise turn results in leftward movement of the hands, which is opposite to the right turn of the wheel. If we look at how people code their responses in this situation (when instructions do not indicate one reference frame or the other), approximately half of them will be using a wheel-based reference frame and the other half a hand-based reference frame (e.g., Guiard, 1983; Proctor, Wang, & Pick, 2004). However, if we instruct people to turn the wheel to the left for one stimulus and to the right for another, everyone reverts to a wheel-based reference frame. For situations in which stimulus and response locations can be coded with respect to multiple reference frames, we can predict what the most compatible mapping will be only if we know what frame will dominate coding.

A closely related phenomenon to the Simon effect is the Stroop effect (Stroop, 1935/1992). People performing the Stroop task name the ink colors of words that spell out color names. For example, the word “green” could be printed in red ink, and a person’s task is to say “red.” The Stroop effect occurs when the word and the ink color conflict. People find it very difficult to say “red” to the word “green” printed in red ink. Stroop interference occurs in many tasks, not all of which involve colors (MacLeod, 1991). The difference between the Stroop and Simon effects is that the Stroop effect seems to arise from conflicting stimulus dimensions, whereas the Simon effect seems to arise from conflicting response dimensions. Explanations of the Stroop effect have tended to focus on response competition, much like explanations of the Simon effect (e.g., De Houwer, 2003).

One last related phenomenon is called the Eriksen flanker effect (Eriksen & Eriksen, 1974), introduced in Chapter 9. People are asked to identify a target stimulus (usually a letter) presented at the point of fixation. On most trials, the target is surrounded on each side (flanked) by irrelevant letters. For example, a person may be asked to press a left key in response to the letter H and a right key in response to the letter S. On a trial, the person may see the letter string “XHX” or “SHS.” Responses to strings of the form “SHS” are slower and less accurate than to strings of the form “XHX.” The explanation for this effect again involves response competition (e.g., Sanders & Lamers, 2002): The letter S activates the right response, and this activation must decrease or be inhibited before the left response can be made.

All three of these effects, the Simon, Stroop, and Eriksen flanker effects, illustrate that irrelevant stimulus attributes can interfere with performance when they conflict with the stimulus and response attributes relevant to the task at hand. They reflect people’s limited ability to selectively attend to relevant task dimensions and to ignore irrelevant ones. The design of display panels and other interface devices should minimize the possibility for interference and conflict between sources of information.

DUAL-TASK AND SEQUENTIAL PERFORMANCE

In most real-world situations, people are required to perform several tasks or task components at once. For example, navigating a vehicle will require the pilot to perform several manual control actions, often at the same time, and also to negotiate obstacles in his or her environment, some of which may appear or disappear unexpectedly. In this situation, several stimuli may require responses in rapid succession. In these more complex tasks, we need to consider how well a person can select and coordinate multiple responses.

PSYCHOLOGICAL REFRACTORY PERIOD EFFECT

How people coordinate several responses to several stimuli at the same time has been studied using a simple laboratory task called the dual-task paradigm. In this paradigm, two stimuli occur in rapid succession, one after the other. Each stimulus requires a different response, usually something simple like a keypress. In this situation, reaction time to the second stimulus becomes longer the closer in time it appears after the first stimulus. As the time between the stimuli becomes longer, reaction time to the second stimulus speeds up, until it almost reaches that obtained when the stimulus is presented alone. This phenomenon was discovered by Telford (1931), who named it the psychological refractory period effect.

Most explanations of the psychological refractory period (PRP) effect have attributed it to a central response-selection bottleneck (Pashler, 1994; Welford, 1952; see Figure 13.11). According to this account, response selection for the second stimulus cannot begin until that for the first task is completed. When the interval between onsets of the two stimuli is short, the response to the first stimulus is being selected and prepared while the second stimulus is being identified. If the duration between the stimuli is short enough, response selection and preparation for the second stimulus may have to wait until the first response is prepared. The response-selection bottleneck model predicts that there should be a linear decrease in reaction time to the second stimulus as the interval between stimulus onsets increases. When the interval duration is long enough that the two response selection processes no longer overlap, no further decrease in response time to the second stimulus should occur. This is approximately what we see in these kinds of tasks.

FIGURE 13.11The sequence of stages in response-selection postponement.

The response-selection bottleneck model also predicts that increasing the difficulty of stimulus identification for the second task (by making the stimulus smaller or more difficult to see) should not have the same effect on response time that increasing the difficulty of response selection will have (Schweickert, 1983). Because there is no bottleneck for stimulus identification, making the second stimulus more difficult to identify should not produce any change in the response time for the second task as long as the increase in stimulus-identification time is no greater than the waiting time at the bottleneck. In contrast, because variables that affect response-selection difficulty for the second task have their influence after the bottleneck, the extent of the refractory effect will not depend on the interval between the two stimulus onsets.

Several studies have confirmed this basic prediction. In one study, people were asked to identify a high- or low-frequency tone by pressing one of two keys with fingers on their left hand (Pashler & Johnston, 1989). After the tone, they were shown a letter (A, B, or C, visually), which was to be classified by pressing a key with one of three fingers on the right hand. This is a standard dual-task paradigm. The experimenters carefully manipulated stimulus identification and response selection difficulty together with the length of the interval between the tone and the letter. The delay between the tone and the letter was either short or long. Stimulus identification was made difficult by reducing the contrast of the letter (gray on a dark background) on half of the trials and by increasing the contrast (white on a dark background) on the other half. Response selection was made easy or difficult by exploiting the fact that it is easier to repeat a response on successive trials than to switch to a new one. On a percentage of the trials, the letter (and therefore the response) was a repetition of the one from the previous trial, and on the remainder, the letter was different from the previous trial.

Figure 13.12 shows that responses to the letter were slower at the short intervals than at the longest one, indicating a PRP effect. More important, as predicted by the response-selection bottleneck model, the PRP effect was larger when letter identification was easy than when it was more difficult (see Figure 13.12a), but the effect size did not vary as a function of whether the letter was repeated from the previous trial (see Figure 13.12b).

FIGURE 13.12Mean reaction times for task 2 as a function of interval between stimulus onsets (stimulus onset asynchrony; SOA) and (a) difficulty of stimulus identification and (b) difficulty of response selection.

FIGURE 13.13Apparatus used by Rosenbaum et al. (1990) (a) with the number of right-handed subjects who used the overhand or underhand grip to bring (b) the white (right) end of the bar to both target discs and (c) the black (left) end of the bar to both target discs.

Although we can find a lot of data that are consistent with the response-selection bottleneck model, we can also find data that are inconsistent. For example, according to the model, response selection for the second task cannot begin until that for the first is completed. This means that performance of the first task should not be affected by variables related to response selection for the second task. But, several experiments demonstrate such backward crosstalk effects (Hommel, 1998; Lien & Proctor, 2000; Ko & Miller, 2014; Logan & Schulkind, 2000). In one, Hommel asked people first to respond to a red or green colored rectangle with a left or right keypress and second to respond to the letter H or S by saying “green” or “red.” When the rectangle and the letter appeared close together in time, people made faster keypresses when the letter response was consistent with the rectangle color than if it was not.

Backward crosstalk effects have led to two alternative models for explaining the response-selection bottleneck. One assumes that response selection uses a limited capacity central resource that can be partially allocated to each task (Navon & Miller, 2002; Tombu & Jolicœur, 2005). The other model assumes that the central resource for response selection is of unlimited capacity, and the bottleneck is strategically created to ensure that the response for the first task is made before that for the second task (Meyer & Kieras, 1997a,b; see Chapter 9).

These two models imply that there should be conditions under which no PRP effect will be obtained. Indeed, when people practice making their responses in any order, the PRP can disappear (Schumacher et al., 2001). Greenwald and Shulman (1973) provided some evidence to suggest that even without any practice, the PRP effect can be eliminated if the two tasks are ideomotor compatible and processing is thus relatively automatic (see earlier). Although the PRP effect is reduced greatly when individuals are practiced at responding in any order or both tasks are ideomotor compatible, it has been difficult to determine whether the bottleneck is indeed being bypassed or eliminated, as Schumacher et al. and Greenwald and Shulman suggest (see, for example, Lien, Proctor, & Allen, 2002; Ruthruff, Johnston, & Van Selst, 2001, for opposing arguments and evidence). The primary message for human factors specialists is that response selection will be slowed when two or more tasks must be performed close together in time, but this slowing can be reduced under certain circumstances.

STIMULUS AND RESPONSE REPETITION

As we just mentioned, reaction times will be faster on a trial in which the stimulus and response are the same as on the preceding trial. This repetition effect will be greatest when the stimulus for the next trial occurs very quickly after the response. The magnitude of the repetition effect is influenced by several other factors as well (Kornblum, 1973): it will get bigger for larger numbers of stimulus-response alternatives than for fewer, and smaller for stimulus-response alternatives with high compatibility. When responses are not repeated, stimulus-response compatibility effects are larger than when they are. In other words, response repetition is most beneficial when response selection is difficult.

Pashler and Baylis (1991) presented evidence that the interaction between repetition and ease of response selection occurs because repetition enables the person to bypass the normal process of response selection when the stimulus-response link already is in an active state. Their results showed a benefit of repetition only when both the stimulus and the response were repeated: No benefit of response repetition alone was observed. In the most extreme case, when the assignment of responses to stimuli was changed on every trial, neither stimulus repetition nor response repetition produced any speed-up in response time. One way to explain this is that people have expectancies about the next stimulus and its associated response that can either help or hinder the response-selection process.

PREFERENCES FOR CONTROLLING ACTIONS

All of the research discussed in this chapter has examined situations in which a person must choose the right response for different stimuli. These situations resemble the interactions an operator might have with display and control panels, where particular buttons must be pushed or switches flipped in response to displayed information. More realistically, there may be many possible ways to manipulate a control device; for example, a knob could be rotated clockwise or counterclockwise. An operator will need to choose an action from these response alternatives that will achieve a specific goal; for example, to increase volume. We know much less about how selection of action is controlled in these circumstances, but research on grip patterns and display–control relationships provides some insight.

GRIP PATTERNS

Grip patterns are the limb movements and finger placements that people use to grasp and manipulate an object (see Chapter 14). Grip patterns are affected by at least two factors (Rosenbaum, Cohen, Meulenbroek, & Vaughan, 2006). The first involves the properties of the object for which a person is reaching, such as size, shape, texture, and distance. For example, when a person reaches out to grasp an object, the grip aperture (the extent to which your hand is open) is directly related to the size of the object (Cuijpers, Smeets, & Brenner, 2004; Jeannerod, 1981). The larger the object is, the larger a person’s grip aperture will be. However, the grip aperture is not affected much by the distance of the object from the person’s hand.

The second factor affecting grip patterns is the intended use of the object. Figure 13.13a shows an experimental apparatus used by Rosenbaum, Marchak, Barnes, Vaughan, Slotta, and Jorgensen (1990). People in this experiment were asked to pick up a horizontal bar and place either its left end or right end on a platform to the left or right of the bar. Everyone who used their right hand used an overhand grip to bring the white (right) end of the bar to either target disc (Figure 13.13b) and an underhand grip to bring the black (left) end to either disc (Figure 13.13c).

Rosenbaum et al. (1990) emphasized the importance of minimizing movement effort in the context of constraints imposed by joint angles and object locations. People apparently selected their initial grip to minimize the awkwardness of the final position they anticipated. A similar study confirmed this: When people had to grasp two bars, one in each hand, and place them in specific target locations, their grips maximized the comfort of their postures at the end of the movement even if the grips had to be different for the two hands (Weigelt, Kunde, & Prinz, 2006). In other words, the goal of achieving final comfortable postures for both hands determines how people carry out the task.

POPULATION STEREOTYPES

Another situation in which selection among controlling actions has been studied involves display–control relationships. In a typical task, people either are asked to demonstrate the most natural relation between a control and a visual indicator (e.g., a needle) or are required to use a control to align the indicator with a particular dial setting. For many types of displays and controls, certain display–control relationships are preferred over others (Loveless, 1962). In the simplest case, consider a horizontal display whose settings are controlled by movements of a parallel, horizontal control stick (see Figure 13.14). It should be clear from our earlier discussion of S-R compatibility that rightward movement of the control stick should cause rightward movement of the indicator, and so on, rather than having rightward responses associated with leftward movements of the indicator. Because most people would intuitively make this association between the stick and the indicator, the association is called a population stereotype.

FIGURE 13.14Horizontal display and control arrangements, with preferred movements indicated.

More interesting is the fact that population stereotypes are found when there is no direct relation between the display and the control. Often, the settings of linear displays are controlled by rotary knobs, as in some car radios. For such situations, four principles act to determine the preferred relationship:

1.Clockwise-to-right or -up principle: A clockwise turn of the control is expected to move a pointer to the right for a horizontal display or upward for a vertical display.

2.Warrick’s principle: When the control is at one side of the display (see Figure 13.15), the pointer should move in the same direction as the side of the control nearest the display.

3 Clockwise-to-increase principle: Clockwise rotation is expected to correspond with an increased reading on the display scale.

4.Scale-side principle: The indicator is expected to move in the same direction as the side of the control that is next to the display’s scale.

FIGURE 13.15Illustration of Warrick’s principle. The stereotypic control movement to produce an upward movement of the display indicator is counterclockwise for (a) and clockwise for (b).

As with Gestalt organization, it is possible to vary the extent to which a particular display–control relationship is consistent with these principles. Hoffman (1990, 1997) evaluated the relative contributions of each of these principles for horizontal displays. Groups of engineers and psychologists indicated their preferred direction of movement for 64 display–control arrangements composed of 8 control locations (see Figure 13.16), 2 directions of scale increase (left, right), 2 types of indicator (a neutral line or a directional arrow), and 2 scale sides (top, bottom). For these situations, the dominant principles were the clockwise-to-right and Warrick’s principles, with the preferred direction of motion predicted well by a weighted sum of the strengths of the individual principles. For engineers, Warrick’s principle was most important, whereas for psychologists, the clockwise-to-right principle was. Hoffman attributed this difference to the engineer’s knowledge of the mechanical linkage between control and pointer, with which Warrick’s principle is consistent. It may be speculated that the engineers’ mental models of the control–display relationship incorporate this knowledge. More generally, the difference between engineers and psychologists illustrates that the population of interest must be considered when evaluating display–control relations.

FIGURE 13.16Locations of the knobs tested by Hoffman (1990).

Another factor that influences expected display–control relations is the orientation of the control operator. Worringham and Beringer (1989) had people guide a cursor to one of 16 target locations with a joystick. The joystick was always operated with the right hand, but the positioning of the arm, hand, and trunk was varied across 11 experimental conditions (see Figure 13.17). With this procedure, the effects of three types of compatibility could be distinguished. Visual field compatibility is display movement that mirrors the control movement while the person looks at the control. Control–display compatibility is defined in terms of the actual direction of movement of the control relative to the display; visual–trunk compatibility arises when the control movement is in the same direction as the display movement relative to the operator’s trunk. Visual field compatibility is the most important determinant of a person’s expectations, regardless of the person’s physical orientation.

FIGURE 13.17Relationship between direction of arm movement and cursor movement, and positions of hand, trunk, and arm, for Worringham and Beringer’s (1989) experimental conditions.

The implication of visual field compatibility being the most important influence on expectations can be understood by considering the condition in which the person’s head is turned to view a moving cursor located on the left side of the body and the control is located on the opposite side, to the right of the trunk. This arrangement has display–control compatibility when the joystick moves forward to move the cursor forward and backward to move it backward, but visual field compatibility when the joystick moves backward to move the cursor forward and forward to move it backward. Yet, the mapping with visual–motor compatibility yielded better performance than the one with display–control compatibility.

Visual field compatibility applies not only to the operation of horizontally moving controls but also to the operation of vertically moving controls and rotary controls (Hoffmann and Chan, 2013). This relation is called the Worringham and Beringer principle. In Hoffman and Chan’s words,

For design purposes, the Worringham and Beringer principle for the relationship between display and control movements, when the operator is moving and the display may not be in the same plane as the control, is the most powerful design principle available for stereotype strength prediction. Designers, where possible, should use this principle in design. (p. 1623)

Up to this point, we have only discussed population stereotypes for two-dimensional movements (left-right, up-down). There are also population stereotypes for complex, three-dimensional displays. One study demonstrated how, for three-dimensional displays, pushing motions were preferred over pulling motions for rightward shifts, backward shifts, and clockwise rotations in the frontal plane (Kaminaka and Egli, 1984). Pulling motions were preferred for upward shifts and rotations toward the operator.

Stereotypic responses also have been demonstrated for controls that are not associated with displays. Hotta, Takahashi, Takahashi, and Kogi (1981) conducted a survey of preferred direction of motion for controls used regularly in daily life. People were shown cubes, each of which had a rotary lever, a slide lever, or a pushbutton on the front, top, bottom, or left or right sides. Given the task of turning a doorknob, turning on water, gas, or electricity, or producing a more generic “output increase” (increasing volume, temperature, speed, etc.), people selected different stereotyped movements shown in Table 13.1. Preferred directions depend on the purpose of a control and the plane in which it is located.

A person’s ability to manipulate controls may be degraded under normal operating conditions when display–control relations are incompatible or inconsistent with population stereotypes. Errors can be minimized by assigning control functions to be consistent with the stereotypes. Table 13.2 summarizes some of the recommended relations between control actions and functions. In emergencies, control actions are more automatic and less deliberate, and stereotypic response tendencies tend to emerge. Loveless (1962) describes a case where the ram of a heavy hydraulic press was raised by pushing a lever down. When an emergency occurred that required the ram to be raised, the ram operator mistakenly made the more stereotypic response of pulling the lever up, which caused the ram to move down and destroy the press. The moral of this story is that it always is best to use display–control relationships that are highly consistent with population stereotypes.

TABLE 13.2

Recommended Control Movements

Control Function	Response Outcome
On	Up, right, forward, pull
Off	Down, left, rearward, push
Right	Clockwise, right
Left	Counterclockwise, left
Up	Up, rearward
Down	Down, forward
Retract	Rearward, pull, counterclockwise, up
Extend	Forward, push, clockwise, down
Increase	Right, up, forward
Decrease	Left, down, rearward

SUMMARY

Response selection is a critical aspect of human performance. The operator of a human–machine system is faced with displays of information that indicate specific actions that he or she needs to take. In many systems, the time with which these response-selection decisions are made is crucial, as is their accuracy. The relative speed and accuracy of responding in a particular situation will be influenced by the threshold used to evaluate the accumulating information. With a high threshold, responses will be slow but accurate, whereas with a low threshold, they will be fast but inaccurate.

The efficiency of response selection is affected by many factors. These include the number of possible stimuli, the number of possible responses, the interrelationships between stimuli and responses, and the amount of practice a person has had in making responses. Moreover, many limitations in the performance of simultaneous multiple tasks can be traced to the response-selection stage. Probably the most important factor in response-selection efficiency is the compatibility of stimuli and responses. Principles of compatibility can be applied to ensure that the easiest or most natural control actions are required in response to displayed information.

When manipulating objects in the environment, the operator has a range of alternative actions for accomplishing a goal. Information about the objects to be grasped and the resulting postures for the limbs is involved in the selection of any particular action. In the next chapter, we will examine the way that actions are controlled.

RECOMMENDED READINGS

Hommel, B., Brown, B. R. E., & Nattkemper, D. (2016). Human action control: From intentions to movements. Switzerland: Springer.

Hommel, B., & Prinz, W. (Eds.) (1997). Theoretical issues in stimulus-response compatibility. Amsterdam: North-Holland.

Newell, A., & Rosenbloom, P. S. (1981). Mechanisms of skill acquisition and the law of practice. In J. R. Anderson (Eds.), Cognitive skills and their acquisition (pp. 1-55). Hillsdale, NJ: Lawrence Erlbaum.

Pachella, R. G. (1974). The interpretation of reaction time in information-processing research. In B. H. Kantowitz (Ed.), Human information processing: Tutorials in performance and cognition (pp. 41-82). Hillsdale, NJ: Lawrence Erlbaum.

Proctor, R. W., & Vu, K.-P. L. (2006). Stimulus-response compatibility principles: Data, theory, and application. Boca Raton, FL: CRC Press.

Sanders, A. F. (1998). Elements of human performance: Reaction processes and attention in human skill. Mahwah, NJ: Lawrence Erlbaum.