Acta Psychologica 36 (1972) 450-458;
0 North-Holland Publishing Company
Not to bereproduced in anyformwithout written permission fromthepublisher
THE EFFECT OF EXPECTATION
ON JUDGMENTS
OF ODOR1
TRYGG ENGEN Injury Control Research Laboratory, Providence, Rhode Island, U.S.A. ABSTRACT
The effect of expectation on the accuracy of odor judgments was studied in two ways intended to simulate factors in real life. First, the relative frequency of the presentation of the odorant versus a nonodorous substance was varied. In addition, the appearance of the substances was manipulated by coloring some of them and not coloring others. The main finding is that both of these variables affect the subjects’ tendency to make the error of reporting the perception of an odor when no odorant has in fact been presented. This tendency persisted even though the subjects were given immediate information ‘feedback’ about the accuracy of each judgment. It seems to reflect a strong and apparently unconscious response bias. The results are discussed in light of contemporary psychophysical detection theory.
1.
INTRODUCTION
The ability of people to perceive and avoid harmful chemical agents with the use of the sense of smell is an important problem in modern society. It could play an important role in avoiding injury and in defining air quality. The validity of such a psychophysical approach, however, is
sometimes limited by the failure of human observers to distinguish between real and imagined odors. As early as 1898, SLOSSON (1899) demonstrated that what he called odor hallucinations are easily obtained and have marked effects on a person’s behavior and his feeling of comfort. In recent studies of the degree of discomfort experienced from exposure to odorous air from pulp mills, it has become evident that there is a strong tendency to report the presence of unpleasant odors when no odorants are actually emitted (CEDERL~~Fet al., 1964; JOHNSSON, 1964; LINDVALL, 1970). The general problem is that people tend to report the experiencing of odors in proportion to their expectation based on nonolfactory cues, such as distance from a mill, black smoke, and other characteristics associated with odors. The present study has to do with the problem of detection, and the likelihood that the subject will make errors of judgment because of his expectation based on the irrelevant appearance (color) of the odorants. 1 The author wishes to thank Mr. M. C. Oviatt for his help with the preparation materials used in this study. 450
of
EXPECTATION ANDJUDGMENTS OF ODOR
451
In addition, research in other sense modalities (GREEN and SWETS, 1966) has demonstrated that the relative frequency of the occurrence of a signal influences the likelihood that the observer will report that a signal is present both when it actually is present (a ‘hit’) and when it is not (a ‘false alarm’). The relationship between the proportions of false alarms and hits for a number of trials as a function of the probability of the occurrence of the signal is described as the receiver operating characteristic (ROC curve). This function has been of indispensable value in practical studies of other sense modalities, and should be useful in the present study of odor perception. 2. 2.1.
METHOD Subjects
One woman and two men, 21 to 26 yr old, were tested extensively. They were obtained from the subject-call file of people responding to an advertisement to serve as subjects in the Injury Control Research Laboratory. The subject was paid $1.60 per hour plus what he earned on the basis of the accuracy of his detection performance based on a payoff matrix. 2.2.
Materials
The experiment was performed in a small room, with temperature of about 22°C and humidity at 70 %. A weak concentration (5, 10, or 20 mM) of n-butyl alcohol in ethyl phthalate was used as the odorant. In agreement with an earlier experiment (SEMB,1968) with the same method of presentation, these concentrations elicited between 0 and 100 % correct detection from an average subject. An attempt was made to obtain a range of accuracy of detection, but individual differences were not the concern of the present study. In order to measure response bias, ethyl phthalate - the diluent - was used as a ‘blank’ and was presented to the subject for observation in the same manner as the odorants. The three subjects reported that this diluent had a weak odor, but they found it discriminable from the odorant in preliminary comparisons. Both the odorant and the diluent were presented to the subject on a cotton swab on a glass rod attached to the cork of a lo- by 75-mm test tube. The cotton was immersed in the liquid odorant and diluent in the test tube when not being used by the subject. The experimenter first demonstrated how to remove the glass rod and sniff the cotton with both
452
T. ENGEN
nostrils in a consistent manner. Then the subject was presented with a test tube on an acrylic plastic holder and was given the following information : This defines what is meant by ‘no odor’ in the present context. The cotton contains a solvent frequently used by perfumers to dilute perfume ingredients. You may judge that it has some odor, but in this context it defines the absence of odor. Attempt to discriminate between it and the content of other test tubes to which I have added a small amount of an odorous substance. It is a weak odor, probably too weak for you to describe its quality, and it is harmless. Just try to detect something different from this pure diluent. The diluent and the odorant will be presented in a completely random order. You can detect it only by smelling. Ignore other characteristics, such as incidental marks on the corks, cotton, or test tubes. If, after smelling the cotton, you think it has an odor, say ‘Yes’; if you think it is the diluent that you smell, say ‘No’.
Any questions? The blank was presented before the beginning of all four experimental conditions, each of which was performed on a separate day, and the subject was encouraged to sniff it as often as he wished in order to become thoroughly familiar with it. There were 10 duplicates of each blank and odorant used in the experiment (40 altogether) in order to prevent the subject from learning the identity of any one test, and they were replaced at least twice for each subject. They were kept out of view from the subject. Only one test tube was presented at a time. There was a 15-set interval between the presentations of test tubes, and a longer rest period of about 5 min after every 50 test tubes. A different random order of presentation for each subject was obtained by using the tables of FISHER and YATES (1963) for random permutations of 10 and 20 numbers. These tables were used in order to obtain a good distribution of events - represented by specific numbers - throughout the whole series, as well as for each block of 50 stimulus presentations. For each experimental condition, a total of 540 tests were presented, of which the first 40 were used for ‘warm-up’ and were not scored. There were four different experimental conditions; each was performed on a separate day and required at least 4 hr. On the first day, the experiment consisted of a test of the effect of using artificially dyed versus clear odorants. This was accomplished by adding the same small amount of Sudan yellow (C17H~N20) to equal amounts of the diluent and the odorant, thereby giving them a yellowish tint without affecting their odors. Thus, on any one trial, the subject might receive a tinted blank, a clear blank, a tinted odorant, or a clear odorant. T’he probability of each event was 0.25. In this part of the experiment, the
EXPECTATION ANDJUDGMENTS OF ODOR
453
subject was not told whether his response was correct or incorrect. The main purpose was to obtain information regarding the hypothesis that the likelihood of a report of odor perception increases with the presence of a potential visual indicator. In the three remaining parts of the experiment the subject won one penny for every correct response and lost one for every incorrect response. The subject recorded gains and losses on separate counters. Except for the payoff matrix, the experimental condition on the second day was the same as on the first day. For the first two conditions of the experiment, the probability of the presentation of a blank was the same as the probability of the presentation of a tinted or a clear odorant, namely, 0.50. For the last experimental conditions (day 3 and 4), this probability was varied, and only clear odorants and blanks were used. The probability that an odorant would be presented was 0.10 on one of those days and 0.90 on the other, with the probability of a blank 0.90 and 0.10 respectively. The order of presentation of these probabilities was decided by the toss of a coin for each subject. He was not given any information about the nature of the experimental conditions other than the correctness of each response. The purpose was to use the odorant-presentation probability to manipulate his expectation that an odorant would be presented and, thereby, increase the likelihood that he would report perceiving odor both on trials with odorants (hits) and on those without odorants (false alarms). 3.
RESULTS
The main results are the proportions of hits and false alarms for the four experimental conditions involving odorant-presentation probability and clear versus tinted odorants. The proportions reported are based on the number of times out of the total possible number that the subject responded, ‘Yes’, to an odorant (hit) and to a blank (false alarm). Since he had to respond either ‘Yes’ or ‘No’, the proportions of ‘No’ for misses and correct rejections may be obtained by subtracting the proportions of hits and false alarms from 1.00. All the results for the three subjects are presented in tables 1 and 2. Although there are individual differences in sensitivity, the results from all subjects are the same with respect to the independent variables of this study. Table 1 shows evidence that tinted sources of odorants are more likely to elicit reports of odor than those that are clear or transparent. Even when the odorant is almost always correctly identified, as in the cases of
454
T. ENGEN
TABLE1 The effects of the color of the odorant and feed-back information on odor detection. Proportions of hits and false alarms obtained by three subjects presented with clear and tinted odorants, millimoles (mM) of n-butylalcohol in ethyl phthalate, and blanks (ethyl phthalate) with and without payoff. Subject 1 (0.01 mM) Subject 2 (0.02mM)
Subject 3 (0.01-0.005 mM)* Clear Tinted
Clear
Tinted
Clear
Tinted
Without payoff Hit False alarm
0.74 0.60
0.74 0.69
1.00 0.62
0.99 0.70
1.00 0.29
0.99 0.71
With payoff Hit False alarm
0.34 0.20
0.53 0.31
0.99 0.04
0.97 0.09
0.86 0.17
0.96 0.26
* The concentration used for this subject was reduced after the first day without payoff in order to make the task more difficult and place his performance in terms of overall correct judgments between S 1 and S 2.
Ss 2 and 3, there is still a greater proportion of false alarms for the tinted than for the clear blanks. For example, S 3 reported 71 % of the time that the tinted blank had an odor, compared with 29 % for the clear blank. It is interesting that none of the subjects ever mentioned this difference in the odorants at any time. After the completion of the experiment, the subjects were asked, ‘Did you notice any color characteristic and, if so, what, if anything, did you think it signified?’ They all responded that they had noticed it but that they had considered it irrelevant. In other words, they seem to have been unconscious of the effect of this variable on their performance. Table 1 also shows that the introduction of the payoff matrix and correction of the judgments had a somewhat different effect on the different subjects, but it did not eliminate the strong tendency to associate odor with the dye. For example, S 1 - who had a relatively difficult discrimination task and made relatively many errors without correction - reduced the total number of responses of ‘Yes’ with the introduction of the payoff matrix, with the result that there were smaller proportions of both hits and false alarms. For S 2, who had an easy discrimination task, the introduction of the payoff matrix had the effect of sharply reducing his proportions of false alarms. For S 3, the task was also made more difficult at the same time by reducing the concentration of the odorants,
EXPECTATION ANDJUDGMENTS OF ODOR
455
and this may have been the reason for the reduction in his proportion of hits. In any case, again the most interesting change is that the proportions of false alarms also decreased although the blanks were, of course, the same as before. This result shows that the introduction of the payoff matrix did not have the expected effect of eliminating the bias related to the visual characteristic of the odorant even after 500 trials. This effect could probably be accomplished with further training (SEMB, 1968; ENGEN, 1960) or with the use of a payoff matrix that fines more heavily for such errors or rewards more for correct judgments, or both, but it is evidently a strong bias. When the probability of the occurrence of the odorant was varied from 0.10 to 0.90 (for clear odorants and blanks only), there was an expected change in the performance of all subjects. Table 2 shows that, as the relative frequency of the odorants was thus increased, the tendency of the subjects to report that they perceived odor also increased. Considered as an isolated fact, this increase in the proportion of hits might suggest improvement in the subject’s sensitivity, but table 2 shows that the proportion of false alarms also increased whenever the proportion of hits increased. This indicates that increasing the relative frequency of occurrence of an odorant will increase the expectation of an odorant and, thus, the tendency to respond ‘Yes’ regardless of the actual situation.
TABLE
2
The effect of odorant-presentation probability on odor detection. Proportions of hits and false alarms obtained by three subjects presented with odorants, millimoles (mM) on n-butyl alcohol in ethyl phthalate, and blanks (ethyl phthalate) with payoff and under three different odorant-presentation probabilities. Odorant presentation 0.10 oso*
probability 0.90
Subject 1 (0.01 mM)
Hit False alarm
0.32 0.04
0.34 0.20
0.97 0.78
Subject 2 (0.02 mM)
Hit False alarm
0.98 0.01
0.99 0.04
0.97 0.24
Subject 3 (0.005 mM)
Hit False alarm
0.82 0.10
0.86 0.17
0.91 0.74
* The results reported in this category are the same as those reported above for clear odorants and blanks under payoff in table 1.
T. ENGEN
456 4.
DISCUSSION
These main findings show that the proportion of hits, as used in classical psychophysical measures of sensitivity, is not an adequate criterion because it is correlated with response bias. They confirm the hypothesis, based on earlier findings, that the presence of a visual indicator increases the tendency people have to report the perception of odor when, presumably, none is present. The proportion of false alarms provides an objective and useful index of this bias. Although it is not easily eliminated, the study shows that this bias can be reduced, if the observer is repeatedly given information that the odor is not correlated with the visual information and if he loses money for judgments made on that presumption. In real-life situations, such feedback is usually not available, and this must then be assumed to affect the interpretation of detection data. The likelihood that the perception of odor will be reported also depends on the observer’s expectation that he will perceive it. The more frequently he has been exposed to an odorant in the past, the more he might report perceiving an odor later, whether or not it actually is present. The proportion of false alarms increases as a monotonic function of frequency of exposure to the odorant. According to contemporary detection theory, sensitivity can be measured independently of the response bias indicated by false alarms by the use of an index, such as d’. This index is computed by correcting the proportion of hits by subtracting the proportion of false alarms after these two proportions have been converted to z-scores or deviations from the mean of a normal distribution with the standard deviation as the unit. It is assumed that the response ‘Yes’ to blanks provides an estimate of the minimum perceptual intensity (odor in this case) spontaneously present in the absence of the experimental odorant and generally described as the random effects of ‘noise’. The effect of presenting a stimulus is assumed to be displacement of the whole sensory distribution of responses obtained to blanks alone (or noise trials) and creating a new distribution called the signal-plus-noise distribution. Both distributions may be assumed to be normal, and the two distributions overlap so that the largest effect from noise alone exceeds the smallest effect of the odorant on the hypothetical sensory dimension forming the common abscissa for these two overlapping distributions. The effect of expectation, as manipulated in the present experiment, is to change the criterion that the observer applies to this dimension in dividing his experiences or perceptions into two groups as required by
EXPECTATIONAND JUDGMENTSOF ODOR
457
the instructions of the experiment; that is, into ‘Yes, I smell something’ and ‘No, I do not smell anything”. The relative locations of the distributions are determined by the effects of random variables as well as by the intensity of the odorant, but these are independent of the subject’s criterion. For that reason, a change in his criterion will affect both the proportion of false alarms - reflecting the noise distribution - and the proportion of hits - reflecting the odorant-plus-noise distribution. If both distributions are normal, and if they are independent of the subject’s criterion, it follows that, by converting them to z-scores, sensitivity can be measured as a difference between these two distributions. This difference, or d’, should be constant regardless of the criterion adopted by the subject. The present data cannot be said to either support or reject this model. Although d’ tended to be constant at values of less than 1.0 for S 1, about 4.0 for S 2, and between 1.0 and 2.0 for S 3 (for 0.005 M), there were considerable intraindividual as well as interindividual differences. There is also the suggestion that, in the present case, d’ tends to be too high for low odorant-presentation probability and too low for high odorant-presentation probability. This is in agreement with the report by subjects that it is easier to discriminate an occasional odorant in preponderance of blanks than vice versa. Considering the short intertrial interval, it is possible that the more frequent presentation of the odorant may have increased the degree of adaptation. The fact that the deviation of d’ from the results predicted by the model is greatest for Ss 2 and 3, who received the perceptually strongest concentrations, may also be consistent with this. As indicated in a previous study (BERGLUND et al., 1970) the existence of such sensory effects would require a modification of the detection model as outlined above. If adaptation affects the signalplus-noise distribution and the receiver-operating-characteristic as indicated by these olfactory data, the model might require an exponential rather than a normal distribution. (BERGLUND et al., 1970). Detection theory has so far mainly considered physical variables and psychological variables associated with response bias, but relatively little attention has been devoted to sensory-physiological variables, such as adaptation, in such contemporary theory. (Accepted
September
5, 1972.)
T. ENGEN
458
REFERENCES BERGLUND,B., U. BERGLUND,T. ENGENand T. LINDVALL, 1971. The effect of adaptation on odor detection. Perception and Psychophysics 9, 435-438. CEDERL~F,R., L. FIUBERG,E. JOHNSSON,L. KAIJ and T. LINDVALL, 1964. Studies of annoyance connected with offensive smell from a sulphate cellulose factory. Nordisk Hygienisk Tidskrift 45, 39-48. ENGEN,T., 1960. Effect of practice and instruction on olfactory thresholds. Perceptual and Motor Skills 10, 195-198. FISHER,R. A. and F. YATES, 1963. StatisticaE tables for medical research,
biological
agriculture,
and
6th ed., New York: Hafner.
GREEN, D. M. and G. A. SWETS, 1966. Signal detection theory and psychophysics. New York: Wiley. JOHNSSON, E., 1964. Annoyance reactions to external environmental factors in different sociological LINDVALL, T.,
1970.
groups. Acta Sociologica 7, 229-263. On sensory evaluation
Nordisk Hygienisk Tidskrif,
of odorous
air pollutant
intensities.
Supplementum 2.
SEMB,G. B., 1968. The detectability of the odor of butanol. Perception and Psychophysics 4, 335-340. SLOSSON,E. E., 1899. A lecture experiment in hallucination. 407-408.
Psychological Review 6,