A Method For Determining the Sequencing of Cognitive Processes in Judgment:&Nbsp; Order Effects on Reaction Times

ABSTRACT - This paper proposes a method for testing hypotheses asserting that to make a judgment, y, the consumer must first make another implicit judgment, x. The method examines reaction times for subjects to explicitly judge both x and y, varying the order and the temporal spacing of the 2 judgments. An illustrative experiment is reported that used the method to study the "false consensus effect"--the tendency for persons to attribute their own opinions and behavior to others. Because of an apparatus problem, faith should not be placed in the data. They are presented only to demonstrate the data analysis procedures employed by the method.

Citation:

John G. Lynch (1981) ,"A Method For Determining the Sequencing of Cognitive Processes in Judgment:&Nbsp; Order Effects on Reaction Times", in NA - Advances in Consumer Research Volume 08, eds. Kent B. Monroe, Ann Abor, MI : Association for Consumer Research, Pages: 134-139.

Advances in Consumer Research Volume 8, 1981      Pages 134-139

A METHOD FOR DETERMINING THE SEQUENCING OF COGNITIVE PROCESSES IN JUDGMENT:  ORDER EFFECTS ON REACTION TIMES

John G. Lynch, University of Florida

[The preparation of this manuscript was supported by the Center for Consumer Research, the Center for Econometrics and Decision Sciences, and the College of Business Administration, University of Florida. Thanks are extended to Ed Shoben for valuable contributions to the conceptualization and design of this research, and to Dipankar Chakravarti and Joel Cohen for helpful comments on earlier drafts of this manuscript.]

ABSTRACT -

This paper proposes a method for testing hypotheses asserting that to make a judgment, y, the consumer must first make another implicit judgment, x. The method examines reaction times for subjects to explicitly judge both x and y, varying the order and the temporal spacing of the 2 judgments. An illustrative experiment is reported that used the method to study the "false consensus effect"--the tendency for persons to attribute their own opinions and behavior to others. Because of an apparatus problem, faith should not be placed in the data. They are presented only to demonstrate the data analysis procedures employed by the method.

TESTING MODELS OF COGNITIVE PROCESS

It is often the case that theories in consumer behavior and related disciplines postulate that one judgment, y, is mediated by another implicit judgment, x. For example, attribution theory (Calder & Burnkrant 1977) postulates that to judge whether a consumer's purchase of a particular brand tells us something about his or her personal dispositions, one must first make a subjective assessment of whether most consumers would have made a similar choice in that situation. However, a judgment of the commonness of choosing a particular brand does not necessarily lead to a subsequent judgment of the personal characteristics of those who buy it.

A question arises as to how a researcher might test a theory that postulates the sort of fixed sequence of mental processes described above (y does not necessarily follow x, but y is necessarily preceded by x). Typically, the best one can do is to marshal evidence that x and y are highly correlated--an unsatisfactory solution to the problem. One could try to improve upon correlation by establishing that x precedes y in time. Unfortunately, if x and y are steps in a sequence of cognitive processes, they may occur too close together in time to be able to easily establish the temporal precedence of x. An alternative approach would be to manipulate x and show predicted effects on y. However, this would show only that x has effect on y. It would not show that a judgment of x is autonomously made whenever y is judged, so that x is a necessary precondition of y. In short, the standard tools of casual inference fail to provide the leverage one needs to test hypothesized relationships between mental events x and y such that a judgment of y implies that one must first make an implicit judgment of x--yet such theories are not at all uncommon in information processing research. This paper will propose a method for dealing with this general class of problems by examining order effects on latency to respond to questions about x and y. To illustrate the application of the method, a flawed experiment is presented that concerns the mental processes underlying judgments of the opinions and preferences of "most people" or "the average person."

Self vs. Other in Consumer Inference-making and Judgment

A pervasive theme in theories of consumer behavior is that a consumer's perceptions and behavior are affected by two broad classes of variables: a) his or her personal beliefs, attitudes, and preferences, and b) perceptions of the beliefs, attitudes, and preferences of generalized and specific others. For example, Burnkrant and Cousineau (1975) contrast informational social influence, in which a person is privately persuaded by the arguments of others, with normative social influences, in which one complies with the judgment of others while remaining privately unconvinced, because of one's concern for how the others would react to one's compliance or noncompliance. A similar self/other duality appears in attribution theory (Calder & Burnkrant 1977), which claims that the perceived cause of a consumer's behavior depends upon how consistent her actions are with the sort of behavior one would expect of most people or the "average person" in a similar situation. Fishbein and Ajzen's (1975) model asserts that a consumer's intention to buy a product is influenced by his personal attitude toward the act of buying the product, and his perception of whether important others think he should or should not make the purchase. Belk's (1976) analysis of gift-giving could be loosely construed as suggesting that when a consumer has no explicit knowledge of the recipient's preferences, he might first decide whether he personally liked the gift, then try to assess whether his tastes and the recipient's were likely to be similar in the product class being considered.

False Consensus

Germane to the discussion above is work by Ross and his colleagues (Ross 1977; Ross, Greene, and House 1977), showing that individuals tend to over-estimate the proportion of others who would make the same behavioral choices they would make (e.g., watch a favorite TV show or prefer a particular brand of coffee), and to underestimate the commonness of choices contrary to their own. Ross et al have dubbed this tendency the "false consensus effect." These researchers placed subjects in a variety of imaginary and real situations in which they were forced to choose between two alternatives, A and B. Subjects were asked to estimate the percentage of persons in the general population who would choose each option and to indicate which choice they themselves would make. Results showed that both persons who chose A and those who chose B believed themselves to be in the majority. Other experiments showed that the false consensus effect extended into the belief domain. Subjects attributed opinions to the "average person" that tended to agree with their own.

Possible Sources of False Consensus

One might be tempted to interpret the false consensus effect as evidence that judgments of what others think are based upon one's own opinions--to judge what the overage person would say, one considers one's own opinion, then generalizes from oneself to the average person. Such a conclusion would be unwarranted, since previous research on the false consensus effect was purely correlational. The correlation between (x) one's own choice of A or B and (y) the perceived commonness of choosing A might, like any correlation, be taken as evidence that a) x caused y, that b) y caused x, c) that both were influenced by some third factor, z, or d) that x and y are part of some bi-directional causal system. Below are some of Ross et al's (1977) speculations about possible sources of false consensus. In the discussion that follows, it should be remembered that "cause" refers to "implicational causation"--x causes y if a judgment of y implies that the person must first activate knowledge structures pertaining to x.

'Independent Judgment" Explanations (z -> x, and z -> y).   A number of explanations of the false consensus effect are consistent with an "independent judgment" model. One such account involves selective exposure. Ross et al note that we tend to associate with friends who share our beliefs and values to a greater extent than these are shared in the general population. Our friends do behave as we do in many instances. The false consensus effect, then, may reflect the fact that our experience leads us to see greater homogeneity in the world than actually exists, rather than any intentional or unintentional distortion of one's social reality to agree with one's own opinions or behavior. This explanation is consistent with an "independent judgment" model because a judgment of one's own opinion and a judgment of the "average person's" opinion activate independent cognitions. A person need not consider his or her own opinion to judge that of the average person, nor does a judgment of one's own opinion necessitate the activation of knowledge structures about what others think.

"Self-referent Inference" Explanations (x -> y).  A second family of causal accounts of the false consensus effect suggests that perceptions of the commonness of various behaviors and opinions are the product of the subject's own choice of behavior or opinion. In general, information about oneself is more extensive and better integrated than information about others, especially if it is tied to an elaborate self-schema (Markus 1977). Self-schemata facilitate both the encoding of new information and the retrieval of information already scored in memory. For these reasons, it may well be that consumers use their own behavior, opinions, etc., as a basis for making judgments of the opinions of generalized and (probably to a lesser extent) of specific others.

"Other-referent Inference" Explanations (y -> x).  A third possibility is that judgments of one's own opinions or preferences are based upon a consideration of what generalized or specific others think. For example, in a choice situation that is fairly novel, a consumer may search for information in memory about what relevant others thought or did in similar situations. Or, if questioned about one's opinion about a topic about which little is known, one may rely on information in memory about what some other person had said about it. Thus, a judgment of one's own opinion might require the activation of cognitions pertaining to what others have said or would say.

A Method for Assessing Implicational Causation

Based upon the evidence presented by Ross et al (1977), any one of the causal mechanisms discussed above might plausibly explain the false consensus effect. Of course, a complete explanation might include several of these mechanisms, and bi-directional causality may be involved. This paper proposes a method for deducing the primary direction of causality underlying the correlation between subjects' choices and their consensus estimates. The method is generally applicable to problems in which two judgments, x and y (in this case, one's own agreement or disagreement with opinion statements and estimates of whether others would agree) are found to be correlated, and one is concerned with whether one is causal in the sense that an implicit judgment of x must be made in order to make a judgment of y. The method involves an analysis of the time it takes a subject to answer questions about x and y as a function of the order in which the questions and the delay between the two questions.

In the experiment to be reported, subjects were shown a series of opinion statements, one by one, on a computer terminal. Each was presented twice during the experiment, a) once preceded by an instruction to judge one's personal opinion (which will be hereafter called a "Self" Judgment) and b) once preceded by an instruction to judge the average University of Illinois student's opinion (which will be called an "Other" judgment). Subjects responded to each statement by pressing either a key marked "AGREE" or one marked "DISAGREE." Let us consider what processes of judgment would be predicted in the various conditions by our 3 causal 'theories," and how reaction times would be affected by these processes.

Components of Reaction Times for 3 Theories

The "self-referent inference" model decomposes the time to make a Self judgment into 2 parts: time to read and Comprehend the opinion statement (designated c) and time to retrieve and integrate knowledge relevant to determining Self's opinion (designated s). [A third component, motor response time to physically execute a keypress, is assumed by all 3 theories to be consistent across experimental conditions, so it will be ignored.] Judgment of Other's opinion is assumed to require both of these same stages (c + s), plus a third stage during which the subject Generalizes from Self to Other (designated g_so). This last stage may involve testing for instances of disagreement, or deciding whether one's opinion about the relevant issue is likely to be typical.

The "other-referent inference" model follows a similar logic. Here a judgment of Other's opinion is assumed to involve a 2-stage process of reading and comprehending the opinion statement (c), and retrieving and integrating knowledge pertaining to Other's opinion (designated o). The Judgment of Self's opinion is assumed to involve a 3-step process: the same 2 stages (c + o) assumed to be required to judge Other's opinion, plus a 3rd stage involving Generalization from Other to Self (g_os).

The "independent judgment" theory concurs with the self-referent theory's account of Self judgments, and with the other-referent theory's account of Other judgments. That is, Self judgments require one to read and comprehend the opinion statement, then retrieve and integrate knowledge relevant to judging Self's opinion (total reaction time = c + s). Other judgments also require 2 stages: a stage entailing the reading and comprehension of the opinion statement, and one involving the retrieval and integration of knowledge pertaining to Other's opinion (reaction time = c + o).

Given the 3 different process accounts of how one judges one's own or the average person's opinion, a method to competitively support one of the models was sought. The problem was attacked by measuring latencies to respond "agree" or "disagree" to opinion statements, a) Subjects made both "Self" and "Other" judgments of each statement, but b) for half the statements, Self judgments were made prior to Other judgments, and the remaining statements were judged in the order Other First, Self Second, and c) the First and Second judgment of each opinion statement were separated by either a Short Delay or a Long Delay. As will be seen, the use of this experimental design with a dependent measure of reaction time permits a clear discrimination among the self-referent, other-referent, and independent Judgment accounts of Self and Other judgments. Following is a discussion of the rationale underlying the "order effects or reaction time" paradigm in the general case.

Logic of the "Order Effects on Reaction Times" Paradigm in the General Case

Suppose that a judgment of y requires an implicit determination of x, plus an additional stage like our Generalization stage, (g_xy), whereas a judgment of x requires only the former stage. If so, judging either x or y at time 1 (First judgment) will facilitate (i.e., reduce the latency of) making the opposite judgment at time 2 (Second judgment, Short Delay) or at time 3 (Second judgment, Long Delay). The facilitating effect results because, in making the Second judgment, the subject need not retrieve and integrate information pertaining to x, but may simply retrieve the integrated determination of x from memory. The average magnitude of the facilitating effect depends upon the likelihood that the subject will be able to (a) successfully retrieve the integrated determination of x, rather than (b) repeat the entire process of retrieving and integrating knowledge about x. The likelihood of being able to follow the shorter path (a) is an interactive function of 2 factors: 1) The delay between the First and Second elements of x-y or y-x pairs. If the delay is short, the integrated determination of x is virtually certain to be in short term memory, enabling the subject to follow path (a). In Long Delay conditions, though, it is less than certain that subjects will be able to retrieve the integrated determination of x from long term memory. 2) In Long Delay conditions, then, the probability of following the shorter path (a) on the Second judgment is positively related to the degree of elaboration of x required by the First judgment. A y-first judgment requires one to judge x, then process x further, whereas an x-first judgment requires no further processing after x has been determined. By a depth of processing rationale (Craik & Lockhart 1972; Craik & Tulving 1975), an x-First judgment should not be as helpful to a y-Second Long Delay judgment as a y-First judgment would be to an x-Second Long Delay judgment. The upshot of this is that the difference between Short and Long Delay versions of an x-Second judgment should be smaller than the difference between Short and Long Delay versions of a y-Second judgment.

Complicating this analysis somewhat is the fact that Second judgments would be expected to be faster than First judgments solely on the basis of practice effects on the speed of the reading and Comprehension (c) stage. This stage would be postulated to he faster for Second Judgments made at time 2 (c₂, occurring in Short Delay conditions) than for First judgments made at time 1 (c_l). By time 3, though, some of the benefit of prior exposure to the opinion statement should have dissipated, so c₃ (in Second judgments, Long Delay conditions) should be of intermediate length (i.e., c₁>c₃>c₂). The assumptions above allow us to derive predictions for the 3 theories of the false consensus effect.

Self-referent Theory.  The self-referent theory's predictions can be summarized briefly by referring to Table 1:

1. A main effect of judgment type should occur. Other judgments should take longer than Self judgments, mainly because the former involve an additional processing step of Generalizing from Self to Other (g_so).

2. A main effect of ordinal position should occur. First judgments of an opinion statement should take longer than Second judgments for two reasons: because of the greater Comprehension time (c) and the greater time to determine Self's opinion (s) associated with First judgments in a (Self-Other or Other-Self) pair.

3. If only Second judgments are analyzed, a significant interaction of opinion Judged (Self vs. Other) and Delay should be observed. Based upon a depth of processing rationale, Other Second judgments should be more adversely affected by a long delay than should Self Second judgments, so in Table 1.

TABLE 1

COMPONENTS OF REACTION TIMES PREDICTED BY "SELF-REFERENT" "OTHER-REFERENT" AND "INDEPENDENT JUDGMENT" MODELS OF IMPLICATIONAL CAUSATION UNDERLYING THE FALSE CONSENSUS EFFECT.

Other-referent Theory.  Based upon a logic similar to that used to derive predictions of the self-referent theory, the following predictions of the other-referent model can be deduced, as shown in Table 1:

1. A main effect of opinion judged should occur. Self judgments should take longer than Other judgments, mainly because the former require an additional step of generalizing from Other to Self (g_os).

2. A main effect of ordinal position is predicted. First judgments of an opinion statement should take longer than Second judgments because of the greater Comprehension time (c) and greater time to determine Other's opinion (o) associated with First judgments.

3. By a depth of processing rationale, one can predict that the difference in RT between Self Second, Long Delay and Self Second, Short Delay cells should be greater than the difference between Other Second, Long Delay and Other Second, Short Delay cells.

Independent Judgment Model. According to the independent judgment theory, knowledge relevant to determining one's own opinion and that relevant to determining Other's opinion exist in separate locations in memory, although both may have been influenced by some common third factor or complex of factors when originally stored. Self judgments are assumed to require subjects to Comprehend the opinion statement (c), then retrieve and integrate knowledge pertaining to Self's opinion (s). Other judgments are assumed to require a Comprehension stage (c), plus a stage involving the retrieval and integration of knowledge pertaining to Other's opinion (o). Retrieving and integrating knowledge relevant to making a Self judgment would not facilitate a later Other judgment, nor would the reverse be true. As shown in Table 1, Second judgments would be faster than First judgments simply because of the shorter time it takes to read and Comprehend (c) the opinion statement the second time around.

1. It is unclear whether a main effect of judgment type should occur. If time to retrieve and integrate information pertaining to Self's opinion (s) is less (or greater) than the time to retrieve and integrate knowledge pertaining to Other's opinion (o), a main effect will result.

2. A main effect of ordinal position will result, solely because of the greater time needed to read and Comprehend (c) the opinion statement during First judgments than during Second judgment.

3. When only Second judgments are analyzed, there should be no interaction between judgment type (Self vs. Other) and Delay. [These assertions assume that knowledge structures pertaining to Self's opinion and those pertaining to Other's opinion are completely independent. It is possible, though, that the basic idea underlying the independent judgment modal is correct--Self judgments require no implicit consideration of Other's opinion, nor is the reverse true--but cognitions about certain facts must be activated whether one is judging one's own opinion or that of the average student. Such a quasi-independent judgment model makes relatively complicated predictions (see Lynch and Shoben, in preparation) not germane to the application of the order effects on reaction times paradigm in most consumer information processing settings. Hence, it will not be discussed further.]

METHOD

Overview

In the present experiment, 66 undergraduate subjects were seated at cathode ray tube (CRT) computer terminals, on which a series of 32 opinion statements was presented. The statements concerned a variety of topics: energy (e.g., "Nuclear power is too dangerous to be a desirable source of energy"), consumer products ("Most breakfast cereals are worthless from a nutritional standpoint"), leisure activities ("It is more enjoyable to read a book than to watch television"), etc. Each statement was presented twice, a) once preceded by an instruction to judge "YOUR PERSONAL OPINION" ("Self" judgment), and b) once by an instruction to judge the "AVERAGE U OF I STUDENT'S OPINION ("Other's judgment), for a total of 64 judgments. Subjects responded to each statement by pressing either a key marked "AGREE" or one marked 'DISAGREE" as quickly as possible while still maintaining a high level of accuracy. Both latency to respond and the response itself were recorded for each trial. Half of the opinion statement pairs were presented in the order Self First, Other Second, and the other half were presented in the order Other First, Self Second. Additionally, the Second judgment in a given pair was separated from the First by either a Short Delay (1-5 trials) or a Long Delay (separated from the First member of the pair by an interpolated task).

The design, then was an Opinion Judged (Self vs. Other) x Order of Judgment (Self First, Other Second vs. Other First, Self Second) x Delay (Short vs. Long) within subjects design, with 8 items randomly nested in each Order x Delay condition. A different random nesting of items was used for each subject. To simplify the exposition of the experiment, it will be treated as an Opinion Judged x Ordinal Position (First vs. Second) x Delay factorial.

Procedure

Subjects, run in groups of 3 to 5, were told that the experiment was concerned with a) whether they personally agreed or disagreed with each of a set of opinion items, and b) whether they thought that the average student at the University of Illinois agreed or disagreed with the same statements. It was explained that 32 opinion statements would be judged once in each condition, for a total of 64 judgments, but that the order of these judgments would be "scrambled."

Each trial was structured so that an instruction phrase flashed onto the screen, telling the subject to judge either "YOUR PERSONAL OPINION" or "AVERAGE U OF I STUDENT'S OPINION" on the upcoming trial. After subjects had read the instruction phrase, they pressed the space bar on their keyboard. This action caused the instruction phrase to be erased, and an opinion statement to be written on the screen. Subjects were told to read the opinion statement as quickly as possible, to determine what they believed to be the correct response, and to press the appropriate key. Subjects were told that not only were their answers being recorded, but also the time it took them to respond. They were instructed to "respond as quickly as possible without sacrificing accuracy." [Cognitive psychologists employing reaction time as a dependent variable typically give subjects similar instructions to discourage them from trading off speed of response for accuracy. In applications in which one can verify the correctness of a subject's responses, it is imperative that differences in mean RT among experimental conditions are not spuriously caused by differential levels of accuracy (Pachella 1974). In the present context, however, there is no objective basis for assessing the "accuracy" of subjects' judgments, nor is it clear what "accuracy" might mean if it could be assessed.]

The experimental session began with 8 practice trials, during which subjects were given feedback about their reaction times. If no questions arose, the practice trials were followed by 48 experimental trials, during which no RT feedback was given. After these trials (all short delay trials, plus the Self First and Other First trials in the long delay condition), subjects performed a 15 minute interpolated task, then completed the final block of 16 long-delay trials, again with RT feedback. Upon completion of the experiment, subjects were thoroughly debriefed and informally questioned about their processing strategies.

Procedural limitations.  The experiment described was flawed by a problem with the apparatus. Although the computer was programmed to generate a different random assignment of opinion statements to experimental conditions for each subject, there is evidence that randomness was not achieved. A chi-square test of association between the 32 opinion statements and experimental conditions was significant at p< .001. Thus, differences in cell means may be attributable to differences in the content of the opinion statements that appeared in those cells. The results of the experiment are presented for illustrative purposes only, and should not be taken to be empirically valid.

RESULTS AND DISCUSSION

Reaction times, with the additive effects of items and answers partialed out, were analyzed in a Subjects x Opinion Judged x Order (Self First, Other Second vs. Other First, Self Second) x Delay x Items (nested in Subjects, Order, and Delay) (66 x 2 x 2 x 2 x 8) completely within subjects analysis of variance. Items as well as subjects were considered random, in keeping with Clark's (1973) contention that linguistic materials must be treated as random. Data will be reported, however, as an Opinion Judged x Ordinal Position (First vs. Second) x Delay completely within S factorial. Mean reaction times in milliseconds are shown in Table 2.

TABLE 2

EFFECTS OF OPINION JUDGED (SELF VS. OTHER), ORDINAL POSITION (FIRST VS. SECOND), AND DELAY (SHORT VS. LONG) ON REACTION TIMES (IN MEETS) TO ANSWER "AGREE" OR "DISAGREE" TO OPINION STATEMENTS.

Overall, Self judgments were faster (M = 3551 resets) than Other judgments (M = 3642 resets), F (1,65) = 8.84 p < .01. Thus the data appear to be consistent with either the self-reference model or the independent judgment model, but not with the other-referent model. Even so, one might ask whether averaging across subjects obscures individual differences in processing strategies. Perhaps some small subgroup of subjects made judgments consistent with an other referent process, using their knowledge of what the average Other might say to make Self judgments. If so, ordinal violations of the self-reference theory's prediction that Self First RTs should be faster than Other First RTs should be associated with ordinal violations of the prediction that Self Second RTs should be faster than those for Other Second judgments. A chi-square test of the independence of these violations in individual subject data revealed no evidence of contingency (X² (1) = 1.36, p > .10). Ordinal patterns in the data, then, suggest that subjects used a common cognitive process, although they apparently differ in the time they devoted to the various stages in this process. Significant Subjects x Ordinal Position (First vs. Second), and Subjects x Ordinal Position x Delay interactions (F (65,1848) = 2.70 and 1.56 (p < .001), respectively), suggest such individual differences. The data discussed up to this point allow the rejection of the other referent theory and seem to support either the self-reference model or the independent judgment model.

Inspection of the data in Table 2 reveals that Other Second RTs show a 158 msec difference between the short and long delay conditions, while Self Second RTs show only a 41 msec increase, consistent with the self-reference, depth of processing theory. A statistical test of this difference requires that only Second Judgments be analyzed in a Subjects x Opinion Judged x Delay x Items (nested in Subjects, Opinion Judged and Delay) ANOVA. The critical interaction does not reach statistical significance, F (1,65) = 2.45 p>.05. If this experiment were not flawed by non-random assign-merit of items to experimental conditions, one would probably prefer the independent Judgment model to the self-referent model based on these data.

METHODOLOGICAL ISSUES IN EMPLOYING THE ORDER EFFECTS PARADIGM

One problem of the "order effects on reaction times" paradigm as proposed is that a causal theory (in this case, self-reference) and an independent Judgment model make different predictions in only one cell out of 8. Given this high variability inherent in reaction times to consumer stimuli, it is difficult to generate the kind of statistical power that is needed to discriminate two theories whose predictions are so similar. This would have been a problem in the experiment reported, had the data been taken seriously.

A second drawback of the paradigm is conceptual. To distinguish a causal theory (like self-reference) from an independent Judgment model, one has to assume a "depth of processing" notion. Cognitive theorists have recently expressed doubts about the usefulness of Craik & Lockhart's (1972) levels of processing construct (cf. Baddeley 1978), and therefore, the use of the construct in the paradigm proposed by this paper may be unwarranted. To generate the predictions of the self-reference theory, it was assumed that deeper processing of one's own opinion occurred in Other First judgments (in which one implicitly judges Self's opinion, then decides whether it would generalize to others) than in Self First judgments (in which one implicitly judges Self's opinion, then overtly reports one's judgment). Lynch and Shoben (in preparation) discuss methods for testing such assumptions.

CONCLUSIONS

Despite the unresolved problems associated with using the "order effects on reaction times" paradigm, it seems uniquely suited to testing process models that postulate that a judgment of y is mediated by an implicit judgment of x. Following are some examples of issues in consumer information processing that could be investigated through the use of the paradigm: a) Fishbein and Ajzen's (1975) assumption that a judgment of behavioral intention implicitly involves the consideration by subjects of whether they think the behavior is good or bad and of whether most people who are important to them think that they should perform the behavior. (b) The assumption of certain "brand-based" choice theories (see Bettman 1979, for examples) that choice from among a set of alternative brands requires overall evaluations of the brands to be made. (c) Hypotheses about the effects of inferential beliefs on evaluations of products (cf. Cohen, Miniard and Dickson 1979).

Researchers who attempt to use this general paradigm will encounter complications that are unique to the particular cognitive process and structure assumptions of the theory being tested. If one exercises care and creativity, though, the general approach of analyzing RTs to answer questions about each stage in a hypothesized chain of mental events as a function of the order of the questions and the delay separating the questions promises to be a powerful aid to the study of consumer information processing.

REFERENCES

Baddeley, A. D. (1978), "The Trouble with Levels: A Reexamination of Craik and Lockhart's Framework for Memory Research," Psychological Review, 85, 139-152.

Belk, R. W. (1976), "It's the Thought that Counts: A Signed Digraph Analysis of Gift-giving," Journal of Consumer Research, 3, 155-162.

Bettman, J. R. (1979), An Information Processing Theory of Consumer Choice, Reading, Mass.: Addison-Wesley.

Burnkrant, R. E. and Cousineau, A. (1975), "Informational and Normative Social Influence in Buyer Behavior," Journal of Consumer Research, 2, 206-215.

Calder, B. J. and Burnkrant, R. E. (1977), "Interpersonal Influence on Consumer Behavior: An Attribution Theory Approach," Journal of Consumer Research, 4, 29-38.

Clark, H. R. (1973), "The Language as Fixed Effect Fallacy: A Critique of Language Statistics in Psychological Research," Journal of Verbal Learning and Verbal Behavior, 12, 335-359

Cohen, J. B., Miniard, P. W., & Dickson, P. W. (1980), "Information Integration: An Information Processing Perspective," in Advances in Consumer Research, Vol. 7, ed., J. C. Olson, Ann Arbor, MI: Association for Consumer Research, 161-170.

Craik, F. I. M. and Lockhart, R. S. (1972), "Levels of Processing: A Framework for Memory Research," Journal of Verbal Learning and Verbal Behavior, 11, 671-684.

Craik, F. I. M. and Tulving, E. (1975), "Depth of Processing and the Retention of Words in Episodic Memory," Journal of Experimental Psychology, 104, 268-294.

Fishbein, M. and Ajzen, I. (1975), Belief Attitude, Intention, and Behavior, Reading, Mass.: Addison-Wesley.

Markus, H. (1977), "Self-schemata and Processing Information about the Self," Journal of Personality and Social Psychology, 35, 63-78.

Pachella, R. (1974), "The Interpretation of Reaction Time in Information-Processing Research," in Human Information Processing: Tutorials in Performance and Cognition, ed. B. Kantowitz, Hillsdale, NJ: Lawrence Erlbaum Associates.

Ross, L. (1977), "The Intuitive Psychologist and His Shortcomings: Distortions in the Attribution Process," in Advances in Experimental Social Psychology, ed. L. Berkowitz, Vol. 10, N.Y.: Academic Press.

Ross, L., Greene, D. and House, P. (1977), "The 'False Consensus Effect:' An Egocentric Bias in Social Perception and Attribution Processes," Journal of Experimental Social Psychology, 13, 279-301.

REFERENCE NOTE

Lynch, J. G. and Shoben, E., "A Causal Analysis of the False Consensus Effect Using Order Effects on Reaction Times." Manuscript in preparation.

----------------------------------------