PERCEPTION OF WEAK STIMULI By Dr. Darrell Butler The girl and the presents The girl sat watching TV. She was alone. Her parents had gone out to dinner and her older brother and sister were out with friends. She was old enough to stay home alone. Besides, soon she would be a year older. Her birthday was just five days away. That is when she remembered that her parents usually hid her birthday presents under their bed. She really wanted to know what they had gotten her. Without thinking about the consequences, she quickly checked the garage and driveway. Her heart was starting to pound in anticipation. No one was there. So she ran upstairs to her parents bedroom. She turned on the lights. She looked under the bed. There was something. She could see some slippers, some cloth in a plastic bag, and some packages that were wrapped in birthday paper. She began to argue with herself. "You really shouldn't be in here. But, I really want to know what they got me. Maybe I could carefully open them and peek then reseal them with tape. No one would ever know. Yes, but what if they them come home and catch me. What would they do?" She listened. Was that a car? She ran to the window and looked out. It wasn't her parents, it was just a car driving down the street. She found a compromise to her internal argument. "I have good ears. I'll listen for them while I check the packages. If I hear them coming, I'll quickly put them away." So she sat down and pulled one of the packages from beneath the bed. Slowly and carefully she removed the tape from one end of the package. Twice she had to stop and run to the window because she thought she heard her parents' car. But she couldn't see any cars. So now she partially unwrapped the package so she could read the label. "Alright." she thought when she found the label. I really wanted this but I didn't think they would get it. She re- wrapped the package and returned it to its hiding place under the bed. She heard something. Oh, it was just a motorcycle and it was going away from the house. She pulled out a second package. She wondered. Could it be? She was really excited. It was hard to believe they would get that too! She was very excited, but she tried to work carefully. She didn't want to rip the paper. This one was wrapped with paper that was easy to tear and she was having trouble. But she finally got it loose. Her heart was pounding as she peeked inside. "Oh it is what I wanted," she exclaimed to herself. She was just starting to calm down when she heard a "Hrrumph" noise coming from the doorway. She looked up and saw her parents. How did they get there? Why hadn't she heard the car? "Oh my" she thought, "am I in trouble now!" You have probably never done anything like this. But you can identify with this girl's situation. Most of her perceptions appear reasonably accurate. She could see that the garage was empty, she could discriminate the objects under the bed, and she could identify the motorcycle and determine that it was moving away. But she thought she heard cars when there weren't any and somehow, she missed the cues that her parents had come home. These are errors of perception. In this section of the chapter, we will consider some explanations as to why we sometimes make such errors. Detection Thresholds Most people believe that there are perceptual thresholds. By this, they mean that a stimulus (e.g., a light, a sound , or a source of heat) must have at least a minimum level of energy or it cannot be detected. This figure shows what could happen in a simple experiment if a person were asked to detect some weak stimuli (light, sounds, or heat) differing only in intensity. In this example, stimuli 1 through 4 are very weak. They are below threshold and cannot be detected. Stimuli 5 through 8 are above threshold. Thus they are perceivable and can be reported. The threshold in this case is between stimuli 4 and 5. Many psychophysicists have adopted a threshold framework to study detection. They refer to the minimum amount of energy needed for detection to occur as the absolute threshold. Psychophysicists developed several different techniques to measure absolute thresholds. One of most important was the method of constant stimuli. The technique can be used with any kind of stimuli, but here let us consider a typical experiment using very quiet tones. The experimenter selects a set of tones that differ only in loudness. The experimenter usually does some preliminary research to verify that some are probably above threshold and some are probably below threshold. The experimenter then presents these tones to a subject one-at-a-time, over-and-over, in random orders. Each time a tone in presented, the experimenter asks the subject to indicate whether he or she heard the tone. The subject answers "yes" or "no." After the stimuli have been presented many times, the experimenter can plot the subject's responses. Usually the experimenter plots the percent of "yes" responses to each stimulus. If the stimuli include the threshold, the plot would look something like this. You can see that the two figures above do not exactly the same. The data does not have a dramatic change that clearly indicates the threshold. Why? Psychophysicists have argued that there are several possible sources or variability. Perhaps the threshold varies slightly from moment to moment. Perhaps there are sources of error in the data. For example, the equipment used to make the sound may not be able create the sounds in the exactly the same way every time. Possibly subjects' physiology is varying somewhat from moment to moment during the experiment. These variations add some noise or variability to the situation. All three of these possibilities would result in "errors" and such errors would be most likely occur when the stimuli was just under or over the threshold. The result is data such as that above. Since the threshold is not typically obvious in the data obtained this way, psychophysicists estimate the threshold. Assuming that the gradual change in the proportion of "yes" responses is due to variation and error, then a good estimate of the threshold is the stimulus level that is reported half the time. The estimate of the threshold has been indicated in the figure above. Using this technique, psychophysicists have determined thresholds for a wide variety of human perceptions. The human visual system is very sensitive. To give you some idea of how sensitive, Galanter (1962) estimates that the visual threshold is about the same as seeing a candle flame from 30 miles on a dark clear night. The frequencies that can be detected by people depend both upon the characteristics of the middle ear and those of the inner ear. Many animals with heads, and auditory systems, much smaller than humans can hear frequencies that are higher than ones we can hear. Typically the highest sounds humans can hear are around 20,000 to 25,000 cycles/second. Dogs can hear frequencies near 30,000 cycles/second. Bats can hear frequencies near 50,000 cycles per second. Many animals with heads, and auditory systems, much larger than humans, for example elephants and whales, can hear frequencies much lower than those that can he heard by humans. Among the stimuli humans can taste, they are especially sensitive to quinine (it tastes bitter). Among stimuli that humans can smell, they are especially sensitive to musk (it smells musky). Returning to the scenario The threshold concept has been useful for studying and communicating human sensitivities to all kinds of energies and chemicals, but can it explain the correct perceptions and various errors made by the girl unwrapping her birthday presents in the scenario provided at the beginning of this section of the chapter? Certainly it makes sense that when stimuli were relatively intense, they were above threshold and thus perceivable and reportable. For example, lights in the garage and the bedroom made it easy for her to see that the garage was empty and a variety of objects under the bed. The motorcycle was loud enough that she could identify it (and knew it wasn't a car) and could tell that it was moving away, probably because it was getting quieter. In this story, numerous stimuli were probably below threshold and therefore were not described. For example, until she got excited, she probably couldn't hear or feel her heart beating. Furthermore, she was probably not distracted by cars many blocks or miles away because the vibrations were too weak by the time they reached her. The real challenge is whether the theory can handle the errors of perception. Can threshold theory explain why the girl heard cars that weren't there? Such misperceptions can occur because people get confused by noises that have similarities to relevant stimuli. For example, her internal noises (such as those caused by her heart beating) could have sounded somewhat like a car pulling into the driveway. The error could have resulted from misattributing the noise to car sounds. Can the theory also explain the other major perception error, missing the parents arrival? One possible explanation is that when she was excited, there were many sources of noise (heart pounding and air moving in and out of the nose) that masked or camouflaged the sounds in her environment. The parents car was just too weak against this background of noise. The major problem with threshold theory account of the perception errors is that noise has opposite effects in the two situations. How can the theory make reasonable predictions about the effect of noise? What makes noise sometimes camouflage sounds and other times it leads to false alarms? Signal Detection Theory You have probably missed stimuli. For example, maybe you have been in the shower very late at night and didn't hear the phone ringing. This kind of perception error is called a miss. It simply refers to not perceiving or reporting a stimulus that was there. Most people agree that such misses could occur, especially if you were not expecting a phone call. Consider another situation. Have you every been in the shower and thought you heard the phone ring, jumped out of the shower, and found that it wasn't? Maybe you were expecting a call about a new job or from a very attractive person you wanted to spend some time with. This kind of perceptual error is not a miss. It concerns deciding that one perceived a stimulus that wasn't really there. This is called a false alarm. Notice that a key difference in these two situations was whether or not you were expecting a call. The idea that expectations can affect perception creates a potential problem interpreting data obtained using the method of constant stimuli. There are two fundamentally different situations that can lead a person to say "yes I detect it" (correctly detecting a stimulus and false alarming on noise) and two different situations that lead a person to say "no"(correctly reporting that no stimulus was there and missing the stimulus). How can we decide if the yes responses obtained using the method are true detections or simply false alarms? This is not a trivial problem. Even if subjects are basically honest, detection experiments often have many ambiguous trials. Subjects can begin to "hear things" because too many trials have gone by in which they didn't hear anything. Just like the person in the shower expecting a phone call, subjects begin to expect that some trials must include the tone. One solution to this problem is to include "catch" trials in the experiment. A catch trial does not contain any stimulus presented by the experimenter. Subjects should say "no" on all catch trials. If subjects say "yes" to any of the catch trials, then it is obvious that something besides stimulus strength is affecting their responses. You may be interested to know that numerous detection experiments have found that subjects commonly say "yes" on some catch trials. One way to explain how honest subjects could produce so many false alarms is a theory called signal detection theory or the theory of signal detectability. This theory was originally proposed by Green and Swets (e.g., Green and Swets, 1966; Swets, 1974; Tanner and Swets, 1954). Unlike threshold theory, Signal Detection Theory describes detection as a result of the independent action of subjects' sensitivity to the stimuli and a cognitive decision that is influenced by factors such as expectation and motivation. Let us consider these two aspects of the theory. First we will examine how noise and stimuli affect sensory activation. Then we will examine the nature of the decision criterion. Signal Detection theory assumes that detection always occurs against a background of "noise," and that sometimes the level of sensory activation caused by the noise can be mistaken for a stimulus. Remember the example of being in the shower and thinking the phone is ringing? According to signal detection theory, the noises created by the water in the shower can sound like a phone ringing. Although many people find the shower example reasonable, they wonder how similar it is to perception experiments. Don't perception researchers make sure it isn't so noisy in their auditory experiments? Yes they do. But according to signal detection theory there is always noise present. Some sources of auditory noise are in the environment. Anything that creates vibrations can be a source of auditory noise. For example, air conditioners, people walking in a building, and wind hitting a building can all create noise. Special rooms can be built to virtually eliminate environmental sources of noise. But even if a person is put in a sound proof room, there will still be noise. A person's heart, lungs, and other organs create internal noises. These noises vary in intensity from moment to moment, but as a result, there is always at least a little noise that produces activity in sensory systems. From the point of view of a subject in an auditory detection experiment, the activity in sensory nervous system due to noise can be misattributed to a stimulus. The variability in noise from moment to moment can result in both misses (not perceiving a stimulus that occurred) and false alarms (misperceiving noise as a stimulus). From an observer's point of view, at any given moment there is some level of activity in virtually every dimension of sensory experience. For example, there is some amount of activity indicating heat level for every area of the skin, there is some amount of activity indicating loudness of every frequency of sound we can hear, and there is some amount of activity indicating the brightness of colors in various areas of the visual field. These levels of activity vary with the level of noise (e.g., internal sources such as heart beats, breathing, or spleen “gooshes” and external stimulus levels that are not part of the stimulus) and the strength of the stimulus. This figure illustrates how the level of activity varies as a function of noise and stimulus presentation. According to the signal detection theory, an observer must make a decision given the level of activity in a sensory domain. In the figure, we have shown a decision criterion adopted by an observer. If the level of activity is above the criterion, the observer reports "yes" there was a stimulus. Sometimes this will be correct. In this figure, correctly indicating that a stimulus was in the environment is called a hit. But sometimes the observer will be wrong; he or she will make a false alarm. If the level of activity is below the criterion, the observer reports "no" there was no stimulus. When the observer has correctly indicated that no stimulus was in the environment, it is called a correct rejection. However, sometimes the observer will be wrong. He or she will miss the stimulus. An observer's sensitivity to a stimulus is indicated by the relative distance between the mean of the noise only and the mean of the signal with noise distributions. In this theory, there is no threshold resulting in a sharp different between detection and no detection. Instead, the distance, or sensitivity, is a continuum. The placement of the decision criterion, that is, the level of stimulation that a perceiver uses to determine whether a stimulus occurred, is determined by cognitive factors such as expectation and motivation. Although we will not go into the mathematics here, the placement of the decision criterion and the sensitivity of the observer are measured independently in Signal Detection Theory. The ability to separate the effects of sensitivity from factors that affect the decision criterion is one of the great strengths of the Signal Detection Theory. Return to the scenario Let us now consider how the theory of signal detectability helps us to understand the correct perceptions and various errors made by the girl unwrapping her birthday presents in the scenario provided at the beginning of this section of the chapter. When stimuli were relatively intense, e.g. the objects illuminated in the lighted garage, they were well above any reasonable decision criteria she could have used and would be reacted to without error. Numerous objects, such as cars several blocks away, probably produced very weak stimuli that were nearly indistinguishable from the noise and clearly below the decision criterion. In a few cases after she lowered her criteria to be more likely to perceive her parents' return, her internal noises (such as those caused by her heart beating) and weak sounds from the outside world were intense enough to be above criteria and thus responded to even though they were just false alarms. However, once the girl began to concentrate on one of the packages, her decision criterion for listening for parents was no longer held low by her. When her parents returned home, the sounds were just too weak to be above criteria, and she missed hearing them until it was too late. Which Theory is Best? Consider the following experiment. Stunkard and Koch (1964) examined how reports of hunger were related to stomach contractions. The study included obese and normal, men and women. Each subject was asked to swallow a stomach balloon. This device allows scientists to measure stomach contractions. During the following four hours, the experimenters occasionally asked the subjects whether they were hungry, experienced feelings of emptiness, or had a desire to eat. Sometimes subjects were asked these questions while the stomach was contracting (stimulus condition) and sometimes they were asked when the stomach was not contracting (catch trials). If one takes a classical threshold approach, the catch trials are ignored in the analysis. From this point of view, obese men had the lowest threshold, normal weight subjects of both genders have intermediate thresholds, and the obese woman had the highest thresholds. In other words, obese women appear to not be sensitive to stomach contractions whereas obese men appear to be very sensitive. However, the Signal Detection Theory provides a very different summary. Responses on both stimulus trials and on catch trials are used to determine both the sensitivity of subjects and their response biases. This analysis indicated that the subjects did not differ in sensitivity to stomach contractions! Apparently, the only difference among them was attributable to other factors. Obese men tended to report that they were hungry almost all of the time thus they had few misses and many false alarms (see Figure S&P.11b). Obese women tended to report that they were not hungry much of time and thus they had few false alarms but many misses. Normal weight subjects tended to report that they were hungry when the stomach was contracting and showed no bias in the ratio of false alarms to misses. So which theory is really right? Although interesting non-manipulation studies such as this one on stomach contractions suggest that are advantages to a theory that separates sensitivity and bias, it doesn't really settle the issue. Other studies have provided stronger evidence. Beecher (1960) reported that placebos (dummy pain relievers) reduce pain in only 3% of subjects in experimental situations. However, they bring relief to 35% of patients in clinical studies! Such differences cannot be due to simple stimulus effects. They must be due changes in non-sensory variables. Of the two theories, only the signal detection theory can easily explain such effects because it gains explanatory power from the placement of the decision criterion. Zwislocki, Maire, Feldman, & Rubin (1958) reported that auditory thresholds dropped with cognitive factors: monetary reward, practice, and increased friendliness with the experimenter. While these results cause problems for threshold theory (because the theory has no simply way to consider such cognitive factors), they are consistent with Signal Detection theory. Subliminal Perception. The term subliminal is derived from the Latin limen, meaning threshold, and sub, meaning below. Subliminal perception has come to refer to two different kinds of situations: Situations involving the perception of weak stimuli (i.e., below the sensory threshold) and situations in which stimulus is strong enough to be easily identified, but aspects or interpretations of the stimulus may not be perceived. Here let us concern ourselves only with cases of extremely weak stimuli. Do you believe that such stimuli can affect your behavior? The two theories we have been discussing, do not agree about subliminal perception. From the point of view of threshold theory, subliminal stimuli should not have an effect on behavior. The kinds of weak stimuli we are considering here are just too weak to exert a measurable influence. Since they are below threshold, they do not affect us. However, from the point of view of signal detection theory, subliminal stimuli may affect behavior. They are weak stimuli that are difficult to discriminate from noise, but they are typically processed by the perception systems and may have some weak affect on behavior. Such effects would probably be difficult to measure and would require sensitive measurement techniques. One kind of experiment that allows scientists to compare these two perspectives is called semantic priming. First, let us consider a semantic priming experiment that does not include any subliminal stimuli. On each trial, two stimuli are presented. The first stimulus (typically a word or a picture) is presented for a short period of time. Then the second stimulus is presented. Subjects are asked to name the second stimulus. Researchers have shown that the time it takes to name the second stimulus is affected by the first stimulus. For example, subjects say the word "doctor" faster if it was preceded by a strongly associated word such as "nurse" or a picture of a nurse than if it was preceded by a weakly or non-associated word such as "chair" or a picture of a ball. This increase in reaction time (RT) is referred to semantic priming. To test whether subliminal stimuli can semantically prime other stimuli, pretesting is done to assure that the time of presentation of the first stimulus is well below the time that would be needed to identify the stimulus. In fact, it is often so short that subjects are unaware that a stimulus was presented. Clearly, such stimuli are subliminal. Unbiased observers would not report them anywhere near 50% of the time. What is important about this research is that different researchers (e.g., Balota, 1983; Hines,1993) have reported that subliminal stimuli do prime! The speedup of naming the second stimulus is smaller than that obtained for supra- threshold stimuli (i.e., those above threshold), but it is reliably obtained. Such results clearly support the signal detection theory. Based on a variety of studies and reasonable arguments, most psychophysicists have come to agree that the Signal Detection Theory is superior to the threshold theory. Yet psychophysicists still do calculate thresholds at times. They do so because it is useful to have a simple method to compare relative sensitivities. But today, when psychophysicists estimate thresholds, they typically include catch trials in their studies and are very careful to design procedures that encourage observers to adopt an unbiased decision criterion. This greatly eliminates concerns that cognitive factors are producing unknown effects on the estimates of thresholds among subjects in the different experiments. However, when cognitive factors cannot be eliminated or cognitive factors are an important variable in a study, psychophysicists prefer to interpret the study using the Signal Detection Theory rather than thresholds.