Information Reacquisition in Sequential Consumer Choice

ABSTRACT - This paper investigates the review (reacquisition) of previously-acquired information, also referred to as backtracking, in a sequential multiattribute choice task. Specifically, it looks at the following four issues: 1) when does backtracking occur? 2) how exhaustive is the review? 3) what is the order in which attributes are reacquired? and 4) does time pressure affect backtracking behavior? Backtracking occurred whenever individuals faced highly undifferentiated alternatives; It was an exhaustive process (over two-third of the information was reacquired); attributes were reacquired in decreasing rank-order; and finally when facing time pressure, individuals spent a smaller proportion of their time backtracking.

Citation:

Gad Saad (1998) ,"Information Reacquisition in Sequential Consumer Choice", in NA - Advances in Consumer Research Volume 25, eds. Joseph W. Alba & J. Wesley Hutchinson, Provo, UT : Association for Consumer Research, Pages: 233-239.

Advances in Consumer Research Volume 25, 1998      Pages 233-239

INFORMATION REACQUISITION IN SEQUENTIAL CONSUMER CHOICE

Gad Saad, Concordia University

[The data and several of the analyses originate from the author's doctoral dissertation while at the Johnson Graduate School of Management, Cornell University. The author is indebted to several institutions for their financial support at various stages of the project including Cornell University (doctoral fellowship), Concordia University (Faculty Research Development Program) and the Social Sciences and Humanities Research Council of Canada. Finally, the research assistanceship of Khalil Khoury is gratefully acknowledged.]

ABSTRACT -

This paper investigates the review (reacquisition) of previously-acquired information, also referred to as backtracking, in a sequential multiattribute choice task. Specifically, it looks at the following four issues: 1) when does backtracking occur? 2) how exhaustive is the review? 3) what is the order in which attributes are reacquired? and 4) does time pressure affect backtracking behavior? Backtracking occurred whenever individuals faced highly undifferentiated alternatives; It was an exhaustive process (over two-third of the information was reacquired); attributes were reacquired in decreasing rank-order; and finally when facing time pressure, individuals spent a smaller proportion of their time backtracking.

In a sequential multiattribute choice, an individual acquires attribute information across the competing alternatives until sufficient cumulative discrimination has been achieved to justify a final choice. Three of the key issues facing an individual in such an environment are determining which attribute information to acquire next if additional information is needed, integrating the newly-acquired information, and deciding when to terminate the search process. Surprisingly, little research has attempted to look at the backtracking stage, namely the review of previously-acquired information. This paper specifically addresses many of the issues inherent in a backtracking episode including: 1) when does backtracking occur? 2) how exhaustive is the review, i.e., how much information is reacquired? 3) what is the oder in which attributes are reacquired? and 4) does time pressure affect backtracking behavior?

Backtracking

Much of the research that has investigated the review of previously-acquired information has dealt with post-decisional processes. Typically, it is argued that following a choice, an individual will experience dissonance and/or regret (e.g., Festinger’s Cognitive Dissonance Theory 1957; Svenson’s Differentiation and Consolidation Theory 1992). As such, these theories propose mechanisms to combat post-decisional anxiety. For example, Svenson and Benthorn (1992) demonstrated that aspects of the most important attributes on which the non-chosen alternative was already inferior, were further depreciated. In other words, subjects went back to the previously acquired information and distorted it in the hope of consolidating their choices. Janis and Mann (1977) argued that distortion mechanisms need not only take place post-decisionally but could also occur during a decision. In addition, they provide an excellent review of some of the early seminal work on post-decisional processes (chapter 12).

Little research has addressed "on-line" backtracking, i.e., whereby an individual reviews previously-acquired information during as opposed to following a decision. The most pertinent studies assess the reacquisition of information. Jacoby, Chestnut and Fisher (1978) found that, within a 35 x 16 IDB, 5% of the searches were of previously-acquired information. Citing some of his prior work using eye-tracking measures, Russo (1978, p. 566) reports a reacquisition rate as high as 56% (i.e., eye fixation of a previously looked at piece of information). Russo and Leclerc (1994) found evidence of refixations using an eye-tracking methodology to investigate choice processes for everyday nondurables. Recently, Lohse and Johnson (1996) compared two process-tracing methods namely Mouselab (informational display board) and Eyegaze (eye-tracking). They found a reacquisition rate of 47% and 69% respectively across the two methodologies. While researchers have investigated the extent to which backtracking occurs in a given decisional episode, there does not exist any research which has specifically investigated when and why "on-line" backtracking takes place, which specific attributes are reviewed and in what order. These issues are all addressed in this paper.

Recently, the sequential-sampling approach has been used to investigate the stopping strategies that individuals apply in multiattribute choices (c.f. Aschenbrenner, Albert and Schmalhofer 1984; Busemeyer and Townsend 1993; Saad 1994; Diederich 1995; Saad and Russo 1996). The model proposes that once some desired threshold of cumulative differentiation between the competing alternatives is reached, the individual will choose the leading alternative. It is argued here that a similar sequential process guides the decision as to whether an individual will backtrack. Specifically, backtracking should take place if some minimal level of cumulative differentiation, h, is not achieved. In other words, there exists a set of backtracking thresholds (below the stopping thresholds), and if the achieved differentiation lies below these, backtracking occurs. See Figure 1 for a pictorial representation. Given that it is difficult to determine where the backtracking thresholds will be a priori, a more testable version of the latter postulate ensues: choices where backtracking occurs will consist of alternatives yielding lower achieved differentiation in comparison to those where backtracking does not occur (H1). Hence, it is posited that backtracking is more likely to occur when an individual is facing undifferentiated alternatives.

Note that the backtracking thresholds do not span the full attribute space. They only become operative at some xth attribute, i.e., an individual needs to be far enough in the process before having and/or needing something to review. While it is difficult to predict the exact value of x, it is expected that backtracking will occur close to the end ofthe search process (H2). One possible explanation albeit a speculative one is that once an individual realizes that the competing alternatives are not very differentiable, he/she will backtrack to review previously-acquired information as a means of minimizing the likelihood of post-decisional regret and/or dissonance. With regards to the contents of backtracking (i.e., which attributes are reviewed and in what order), it is proposed here that a subset of the most important attributes will be reviewed, in decreasing order of importance (H3). Finally, when facing time pressure, individuals should reduce the proportion of time that they spend backtracking (H4). In other words, it is proposed that under time pressure, subjects will spend a greater proportion of their time either deciding which attribute information to acquire next (attribute selection stage) or integrating newly-acquired information (integration stage).

FIGURE 1

THE BACKTRACKING AND STOPPING THRESHOLDS IN THE DISCRIMINATIONS FRAMEWORK

METHOD

Task

The task consisted of choosing between pairs of apartments to rent for one year. The apartments were defined by 25 attributes. Apartments were used because they represented familiar and consequential decisions for the subject population that was used (university students). For a given pair of apartments, a subject could request anywhere from 1 to 25 pieces of attribute information prior to making a choice. If additional information was desired, the subject could choose which attribute to acquire next. A request for an additional piece of attribute information implied that both attribute values corresponding to the two competing alternatives would appear simultaneously. Subjects participated in four separate sessions. In each session, they made 15 binary choices between competing apartments.

Apparatus

The SMAC computer interface (Saad 1996) was used to implement the sequential decision making environment. The latter collects an impressive set of process-tracing data in each of the distint stages of a sequential choice, namely the attribute selection, information integration and backtracking stages.

Subjects

Twenty-two (17 female) undergraduate students participated in the experiment. They were recruited on the campus of a large northeastern university. Each subject came for 4 sessions, none lasting longer than 1 hour. Subjects were paid $20 for the entire experiment.

Procedure

Subjects’ attribute rankings and weights were elicited using a Q-Sort procedure (see Saad and Russo (1996) for a discussion of the technique). Subsequently, subjects made 15 binary choices between competing apartments. For each of the 15 choices, subjects acquired attribute information one piece at a time, entered the change in their cumulative confidence as a result of this new information and decided whether to stop and choose the leading apartment or to acquire additional information. A cumulative confidence of p in favor of an alternative meant that based on the information acquired so far, there was a (1-p) probability that the preference would be reversed in light of all possible information. The lower boundary of this measure was 50 (toss-up between a pair of apartments), while the upper boundary was 100 (zero chance of a preference reversal, i.e., the currently leading apartment not being ultimately chosen). After each attribute, subjects also had the option of reviewing previously acquired information. They could request to review information on specific attributes and/or view a pictorial record of the cumulative confidence measure up to the current point in the decision (i.e., a tracking curve of the confidence measure).

The same procedure was repeated in each of the four sessions, with the only exception being that in the fourth session, subjects were placed under time pressure. Specifically, they were provided with 65% of the time that they had taken to make the 15 choices in the third session. Thus, the manipulation was subject-specific in that it took into account baseline time differences across individuals. The 65% level was based on the results of a pilot study.

Results

The data corresponding to a subject’s first choice were removed from all analyses to be reported in this paper, in order to account for practice effects. Whenever a subject backtracked on a given trial, the interface collected the following information: 1) the achieved cumulative confidence up to that point (excluding any entered changes while viewing the current attribute); 2) the proportion of the decision completed when backtracking occurred. For example, if in a given trial, the subject backtracked on the fifth acquired attribute and the final number of attributes acquired on that trial was ten, then 50% of the decision was completed when the review occurred; 3) the ranks of the attributes requested for review and the order in which they were reviewed; 4) a "flagging" each time a request was made to review an attribute which had not been acquired in the current trial; In addition, the attribute’s rank was recorded; 5) a "flagging" whenever a request was made to review the pictorial tracking of the cumulative confidence measure.

(i) When and Why Does Backtracking Occur? A within-subjects analysis of the backtracking data is impossible due to the small number of observations per subject in a given session. As such, the data of all 22 subjects were pooled, in each of the four sessions. Recall that H1 posited that backtracking will take place when the achieved discrimination/confidence is low. In addition, it was proposed that reviews would typically occur close to the point at which a decision would be made (H2). Table 1 displays the averages for the achieved cumulative confidence and the proportion of the decision completed when backtracking occurred. The means for the latter two measures across all 4 sessions were 62.7 and 81.6% respectively. The number of observations listed at the bottom of Table 1 correspond to the number of times that a backtracking episode occurred in each of the 4 sessions.

TABLE 1

WHEN DOES BACKTRACKING OCCUR? MEANS OF TWO "TEMPORAL" MEASURES ACROSS THE 4 SESSIONS

There does not exist a benchmark against which 81.6% can be compared. As such, no formal statistical tests can be conducted to gauge the significance of the latter result. However, one can qualitatively conclude that subjects tend to backtrack when they are close to making a choice.

Note that while there were 238 backtracking episodes across the 4 sessions (i.e., the sum of the observations in Table 1), 31 of the backtrackings did not involve any attribute reviews. In the latter cases, subjects simply requested to view the pictorial tracking of the cumulative confidence measure up to that point in the decision. Specifically, there were 9, 15, 7 and 0 such cases across sessions 1 through 4 respectively. These backtracking episodes will be hereafter referred to as the Tracking Curve Only (TCO) cases. As might be expected, none of the subjects displayed TCO backtracking behavior in the time pressure session (i.e., session 4). In looking at the "proportion of the decision completed when backtracking occurred" for the 31 TCOs, an interesting pattern emerged. The means for the latter measure in each of the first 3 sessions (i.e., recall that TCOs did not occur in session 4) were 92.6%, 98.2% and 96.0% respectively. It would appear that TCO backtracking behavior is used as a final check immediately prior to making a final choice between the competing alternatives.

Given that backtracking occurred close to the point at which the final chice was made, one would expect that the average change in cumulative confidence between the point at which backtracking occurred and the final cumulative confidence achieved (i.e., when the choice was made) would be small. The average difference in cumulative confidence achieved was 2.75 across the 4 sessions. As expected, the final cumulative confidence (i.e., when final choices were made) was greater than the achieved cumulative confidence when backtracking occurred, in each of the 4 sessions. Table 2 displays the means of final number of attributes acquired and final cumulative confidence achieved in each of the 4 sessions, for two sets of trials namely, those where backtracking did and did not occur.

To test H1, one needs to determine whether there were significant differences in terms of final cumulative confidence achieved and final number of attributes acquired between the two sets of trials. Clearly, trials were backtracking did not occur should yield larger final cumulative confidence scores (i.e., the competing alternatives are more differentiated). This in turn would imply that the final number of attributes acquired for the latter trials will be smaller than for those trials where backtracking took place. In other words, as postulated by H1, backtracking is expected to occur when choices yield competing alternatives that are not very differentiated. This will manifest itself via the inherent differences in final cumulative confidence achieved and final number of attributes acquired between the two sets of trials. Eight one-tailed t-tests (using the more conservative assumption of unequal variance) were conducted, corresponding to the 8 differences of means in Table 2. The p-values were as follows: 6 (p=0.00), 1 (p<0.02) and 1 (p<0.005), all in the predicted directions. Thus, H1 is strongly supported, namely backtracking occurs when competing alternatives yield low differentiation.

An additional set of exploratory analyses were performed to test whether the backtracking thresholds were constant (See Figure 1). Four linear regressions were performed, corresponding to the pooled data in each of the 4 sessions. The dependent and independent variables were the cumulative confidence achieved and the number of attributes acquired respectively, at the moment when backtracking occurred. None of the four estimated slopes significantly differed from 0, thus demonstrating that the backtracking thresholds are indeed constant throughout the attribute space, as shown in Figure 1.

(ii) Content of Backtracking. The ensuing section discusses the number of attributes that were reviewed and the order in which they were reacquired during the backtracking episodes. The 31 TCO cases were not included in this analysis for by definition no attribute reviews took place, leaving 207 usable backtracking episodes. [If a subject inadvertently "clicked" two or more times on an attribute for review, only the first "click" was counted. Similarly, all requests to review attributes which were yet to be acquired in a current trial were removed from the data set. Requests to review yet-to-be acquired attributes probably occurred because subjects might have confused the information which had been acquired in a previous trial with that of the current one.] The data were once again pooled across subjects for each of the 4 sessions. The average number of attributes reviewed across the 3 no-time pressure sessions was 5.44 while for the time pressure condition, it was 3.47. The percent of previously-acquired information which was reviewed in a given backtracking episode across the 3 no-time pressure sessions was 69.01 while for the time pressure condition, it was 70.14 (difference is ns). Table 3 displays the means for both of the latter two measures across each of the 4 sessions. Thus, whenever backtracking occurred, it was a relatively thorough process, regardless of whether time pressure was present or not.

TABLE 2

MEANS OF THE FINAL CUMULATIVE CONFIDENCE ACHIEVED AND FINAL NUMBER OF ATTRIBUTES ACQUIRED FOR BOTH BACKTRACKING TRIALS (B) AND NON-BACKTRACKING (NB) TRIALS

TABLE 3

HOW EXHAUSTIVE IS THE BACKTRACKING EFFORT?

While the diligent nature of backtracking has been demonstrated, it is thus far unclear what the reacquisition order of the reviewed attributes is. Recall that H3 predicts that attributes are reviewed in decreasing rank-order.

A perfect validation of the latter postulate would imply that if x attributes were reacquired in some given choice, the reacquisition order should be R1, R2,...,Rx, where Ri corresponds to a subject’s ith most important attribute (i.e., Ri=i, i=1,...,x). Thus, in the above example, the difference between the expected and actual ranks would be zero for each of the x reacquired attributes. Similarly, suppose that a subject had the folowing reacquisition order: O1, O2,..., Ox, where Oj (j=1,...,x) corresponds to the subjects’s ranking of the jth reacquired attribute. One can calculate the absolute value of (Ri-Oi), for i=1,...,x, i.e., the deviation between the expected and observed rank for each of the x reacquired attributes. For each reacquired attribute in a given session, the deviation between the expected and actual rank was calculated. Subsequently, to formally test H3, the observed ranks were regressed on the expected ones using the pooled data from each of the 4 sessions. In other words, an estimate of a and b were calculated for the regression equation O=a + bR in each of the 4 sessions. Clearly, a perfect validation of H3 would yield a straight line, which passes through the origin and makes a 45 degrees angle with both the x and y axes respectively (i.e., a=0 and b=1). As such, 4 t-tests of b=1 were performed with all four failing to reject the null hypothesis that b=1 (p<0.05). In addition, an MAVD (i.e., Median Absolute Value Deviation between expected and actual rank) was calculated for each of the 4 sessions. Table 4 displays the relevant measures from the latter analyses.

TABLE 4

MAVDS, R-SQUARED, BETAS (SLOPES) AND ALPHAS (INTERCEPTS) VALUES DURING THE BACKTRACKING STAGE

TABLE 5

PERCENT OF TIME SPENT IN THE ATTRIBUTE SELECTION, INFORMATION INTEGRATION AND BACKTRACKING STAGES

The pattern of results provide strong support for H3, namely that attribute ranks are a very good predictor of reacquisition order. One caveat worth mentioning here is that whenever subjects wished to review previously-acquired information, the attributes were shown in the subjects’ ranked-order. As such, the display format might have elicited the subjects to simply "go down" the rank-ordered list. Given that the regression fits are imperfect, there is no reason to suspect that the latter behavior was prevalent. [A study was recently completed wherein the listing format in the backtracking screen was randomized. A preliminary analysis showed that while the fit between the expected and realized ranks did worsen (beta = 0.50 and R-Squared =20%), it was sufficiently good that one can still safely conclude that attribute ranks are a good predictor of reacquisition order.]

The fact that the display format might have had an effect on subsequent behavior, is consistent with previous research in other information processing contexts (cf., Russo 1977; Todd and Benbasat 1991).

Effects of Time Pressure

H4 postulated that time pressure would alter the allocation of one’s time to each of the three decisional stages within a sequential environment, namely individuals will spend less time in the backtracking stage and accordingly more time in either the attribute selection or integration stages. Table 5 displays the relative average percent of time spent on each of the 3 stages across subjects, for each of the 4 sessions.

As expected, there was great consistency in time allocation across stages in the first 3 sessions, since all were run under no-time pressure. However, it appears that in session 4, subjects reduced the relative time they spent backtracking and accordingly increased the time spent integrating the acquired information.

To test the statistical significance of the change in time allocation across the no-time and time pressure conditions, only the session 3 data were used for the former condition. The latter were the most stable data set among those of the first three sessions, in that it was less likely that learning and practice effects were still occurring. To illustrate this point, the average time spent making the 14 non-practice choices was calculated across subjects, for each of the 3 sessions. Clearly, if the average time decreases monotonically, this would demonstrate a learning effect vis-a-vis the experimental task. As expected, the average time across the 3 sessions was 28.84, [Subject 14 inadvertently quit session 1 at the end of her 7th trial. Using the time it took her to complete the latter 7 trials, a linear extrapolation was used to estimate how long the full session would have taken her to complete. The extrapolated time was used in the above calculation. In addition, parts of subject 3's session 1 data were lost due to a faulty diskette. Luckily, the recovered parts included the relevant time measures. As such, here data was included in the above calculation.] 27.06 and 25.21 minutes respectively, hence illustrating subjects’ increased facility with the task at hand. Clearly, using the data from the first three sessions would have increased the sample size of the no-time pressure condition. Nonetheless, it was felt that the drawback of using a smaller sample size was offset by the increased stability of using data that were unlikely to be affected by learning effects.

The data in Table 5 contain a dependency between the three variables (the 3 means in a given session sum to 1). As such, the statistical test that was used simultaneously compared only 2 out of the 3 means. In other words, based on the results of the latter tst, the corresponding result for the third mean was subsequently inferred. In addition, the statistical test assumes that the data ranges across the whole real line while the raw data here consists of proportions ranging from 0 to 1. Hence, a log(p/(1-p)) transformation (Mosteller and Tukey 1977) was performed on the data, with the transformed data now spanning [-$, +$].

The multivariate paired comparison T²-test (Johnson and Wichern 1988, p. 211-215) was used. The latter is a straightforward multivariate generalization of the univariate matched (i.e., paired) t-test. Using the data from the attribute selection and information integration stages, a statistically significant difference between these two means was found (T²= 15.39; F_2,20 (0.05)=7.33). Simultaneous confidence intervals revealed that the difference was solely due to the differences in the information integration means (larger in session 4). In other words, there was no statistically significant difference between the attribute selection means. Accordingly, given the dependency of the 3 variables, this implied that there was a corresponding statistically significant difference between the backtracking means (smaller in session 4), a pattern in support of H4. Thus, under time pressure, inviduals spend less time backtracking and more time integrating the acquired information. Surprisingly, the proportion of time spent deciding which attribute to acquire next was unaffected by time pressure.

The combined results of Tables 3 and 5 yield an interesting finding. Recall that subjects reacquired the same proportion of information in both the time pressure and no-time pressure conditions (see Table 3) while reducing the percentage of time spent backtracking (see Table 5). This suggests that subjects potentially increased their rate of information processing. The current data does not permit for a direct test of the latter postulate. However, an analysis of some peripheral data not reported here reveals that acceleration did occur (a 35 % decrease in the time available to complete the task yielded 27% fewer information being acquired). [I am indebted to an anonymous reviewer for drawing my attention to this point.]

Summary of the Results

H1 posited that backtracking would take place whenever an individual faced alternatives that yielded low differentiation, while H2 proposed that backtracking would occur close to the end of the search process. Strong support was found for both hypotheses. In addition, an exploratory analysis revealed that backtracking was an exhaustive process (i.e., a large proportion of the acquired information was reviewed), providing indirect support that the latter occurs when individuals are torn between pairs of undifferentiated alternatives. In terms of the content of backtracking, attributes were reacquired in decreasing order of importance, as predicted by H3. Finally, when facing time pressure, subjects did not backtrack as often, supporting H4. Instead, they spent relatively more time integrating the acquired information.

GENERAL DISCUSSION

In the current interface, a subject could solely request to review previously-acquired information, albeit in an online manner (i.e., prior to a final choice being made). An updated version of the interface will permit subjects not only to review previously-acquired information but also to modify its impact. In his differentiation and consolidation theory, Svenson (1992) argued that individuals apply post-decisional consolidation processes to defend the chosen alternative against possible threats. Two such mechanisms include bolstering (discounting) the impact of attributes that the chosen alternative is superior (inferior) on. By providing subjects with the opportunity to alter the impact of previously-acquired attributes during the backtracking stage, this would permit for an online (i.e., prior to a final choice being made) test of Svenson’s theory. In recent paper, Russo, Medvec and Meloy (1996) found evidence of predecisional distortions of newly-acquired information. The impact of the information favoring preferred (leading) alternatives was distorted as to further increase the differentiation between the preferred and non-preferred alternatives. An interesting area for future research would be to investigate whether such distortions occur during an online backtracking episode (i.e., distorting the impact of previously-acquired information upon further review). Assuming that the acquisition costs of the yet-to-be acquired information are smaller than those of previously-acquired information, individuals might be more likely to distort the impact of the latter more often than that of the former. Lowenstein (1996) recently argued that much of the tradition in behavioral decision theory has ignored "hot" cognitions such as emotions and passions. Motivational distortions of the objective impact of acquired information, whether arising predecisionally and/or post decisionally, might be due to such "non-rational" factors as those identified by Lowenstein. For example, brand loyal consumers might be sufficiently passionate vis-a-vis their preferred brand, as to engage in a motivational distortion of information to support their preference. This is not unlike two sports fan, passionate in their support for opposing teams, who upon a television replay (i.e., backtracking), will arrive at divergent opinions about the veridicality of the referee’s decision.

The current research along with recent work by Saad (1994) and Saad and Russo (1996) are a first step towards integrating the several stages of a sequential choice process under a common discrimination framework. Whether deciding which attribute to acquire next (acquisition strategy), which attribute to reacquire (backtracking strategy) or when to stop acquiring additional information (stopping strategy), the discrimination framework posits that these decisions will be motivated by an individual’s desire to maximally discriminate/differentiate between the competing alternatives. In light of the framework’s ability to explain a myriad of behaviors at several stages of the sequential choice process, the potential for future research in this area appears fruitful.

REFERENCES

Aschenbrenner, K.M., Albert, D. & Schmalhofer, F. (1984). Stochastic Choice Heuristics. Acta Psychologica, 56, 153-166.

Busemeyer, J. R. & Townsend, J. T. (1993). Decision Field Theory: A Dynamic-Cognitive Approach to Decision Making in an Uncertain Environment. Psychological Review, 100, 432-459.

Diederich, A. (1995). A Dynamic Model for Multi-Attribute Decision Problems. In Contributions to Decision Making-I, J. -P. Caverni, M. Bar-Hillel, F. H. Barron and H. Jungermann (Eds.), Elsevier Science B.V., 175-191.

Festinger, L. (1957) A Theory of Cognitive Dissonance. Stanford, CA: Stanford University Press.

Jacoby, J., Chestnut, R. W. & Fisher, W. A. (1978). A Behavioral Process Approach to Information Acquisition in Nondurable Purchasing. Journal of Marketing Research, 15, 532-544.

Janis, I. L. & Mann, L. (1977). Decision Making: A Psychological Analysis of Conflict, Choice, and Commitment.. New York: The Free Press.

Johnson, R. A. & Wichern, D. W. (1988). Applied Multivariate Statistical Analysis. New Jersey: Prentice Hall.

Loewenstein, G. (1996). Out of Control: Visceral Influences on Behavior. Organizational Behavior and Human Decision Processes, 65, 272-292.

Lohse, G. L. & Johnson, E. J. (1996). A Comparison of Two Pocess-Tracing Methods for Choice Tasks. Organizational Behavior and Human Decision Processes, 68, 28-43.

Mosteller, F.& Tukey, J. W. (1977). Data Analysis and Regression: A Second Course in Statistics. Reading: Addison-Wesley Publishing.

Russo, J. E. (1977). The Value of Unit Price Information. Journal of Marketing Research, 14, 193-201.

Russo, J. E. (1978). Eye Fixations Can Save the World: A Critical Evaluation and a Comparison Between Eye Fixations and Other Information Processing Methodologies. In Advances in Consumer Research, H. K. Hunt (ed.), Ann Arbor Association for Consumer Research, 561-570.

Russo, J. E. & Leclerc, F. (1994). An Eye-Fixation Analysis of Choice Processes for Consumer Durables. Journal of Consumer Research, 21, 274-290.

Russo, J. E., Medvec, V. H. & Meloy, M. (1996). The Distortion of Information during Decisions. Organizational Behavior and Human Decision Processes, 66, 102-110.

Saad, G. (1994). The Adaptive Use of Stopping Policies in Sequential Consumer Choice. Unpublished doctoral dissertation, Johnson Graduate School of Management, Cornell University.

Saad, G. (1996). SMAC: An Interface for Investigating Sequential Multiattribute Choices.Behavior Research Methods, Instruments, & Computers, Vol. 28 (2), 259-264.

Saad, G. & Russo, J. E. (1996). Stopping Criteria in Sequential Choice. Organizational Behavior and Human Decision Processes, 67, 258-270.

Svenson, O. (1992). Differentiation and Consolidation Theory of Human Decision Making: A Frame of Reference for the Study of Pre- and Post-Decision Processes. Acta Psychologica, 80, 143-168.

Svenson, O. & Benthorn, L. J. (1992). Consolidation Processes in Decision Making: Post-Decision Changes in Attractiveness of Alternatives. Journal of Economic Psychology, 13, 315-327.

Todd, P. & Benbasat, I. (1991). An Experimental Investigation of the Impact of Computer Based Decision Aids on Decision Making Strategies. Information Systems Research, 2, 87-115.

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