Decision Process Theory and Research

Peter Wright, Stanford University
ABSTRACT - This paper discusses the theorizing, research designs and data presentations in three studies of decision processes. The focus is on the development of theoretical propositions consistent with prior research and on research strategies for testing such propositions.
[ to cite ]:
Peter Wright (1979) ,"Decision Process Theory and Research", in NA - Advances in Consumer Research Volume 06, eds. William L. Wilkie, Ann Abor, MI : Association for Consumer Research, Pages: 218-221.

Advances in Consumer Research Volume 6, 1979      Pages 218-221


Peter Wright, Stanford University


This paper discusses the theorizing, research designs and data presentations in three studies of decision processes. The focus is on the development of theoretical propositions consistent with prior research and on research strategies for testing such propositions.


Each of the three papers to he reviewed (Walton and Berkowitz, 1978; Raju, 1978) presents propositions about aspects of the decision-making process, and two papers present data. The theoretical propositions are discussed in terms of their clarity and consistency with earlier re -search. In discussing the empirical tests, the chosen research strategy is discussed in terms of alternate strategies one might use for researching the same question. The intent of this discussion is to position these three studies relative to the larger stream of research of which they are a part.

Walton and Berkowitz: The Effects of Choice Complexity and Decision Freedom on Consumer Choice Behavior

The Walton and Berkowitz experiment studied the relationship between two antecedent factors--- the number of available alternatives (Nalt) and the amount of each alternative available for consumption--and three behaviors---total time to decide one's preference, post-decisional feelings about one's earlier decision freedom, and actual consumption of the chosen alternative. The experiment's motivation is commendable. Variables which theorizing suggests may play roles in conditioning decision processes were selected, and a study that represents both a replication and an extension of prior experiments was conducted. Similarly, the design of the experiment seems free of obvious artifacts, as best we can tell from the report, and includes one procedural feature---the equating of the attractiveness of the stimulus alternatives presented to each subject---which helps clarify the Nalt manipulation.

Relationship to Prior Research

The presentation of the data might be more enlightening if it were accompanied by more detailed theorizing about hypothesized relationships and more direct interpretations of the disparity between the Walton and Berkowitz results and the earlier Reibstein et al (1975) results. Let me briefly discuss the disparity issue first.

Reibstein et al found that subjects who chose between four soft drinks reported having felt more decision freedom and drank more of the chosen alternative than subjects who chose between two drinks. Walton and Berkowitz note that the relative attractiveness of the four drinks in the Reibstein et al study was not scaled a priori, nor equated at the individual subject's level. They correct this in their study, and hint that it is a crucial design feature, yet do not explain why explicitly. It seems desirable to make explicit why this is important in interpreting the observed disparity in findings; in so doing, I am presumably only stating what Walton and Berkowitz had in mind but did not state.

There is a greater probability that a set of four randomly chosen drinks will include a drink the subject finds especially attractive than if only two of the four are offered. Hence, with no prior scaling of the drinks' attractiveness to each subject, we would expect more of Reibstein et al's "four alternative" subjects to find a highly attractive drink in their set than is true of the "two alternative" subjects. If we then merely assume that the amount someone drinks depends on how much he likes what he is drinking, we would predict greater consumption of the chosen drink in the case where Nalt =4, just as Reibstein et al observed. This consumption behavior would not depend in any way on perceived choice freedom, and the interpretation in that vein offered by Reibstein et al would be questionable. Walton and Berkowitz eliminated this design artifact, and found no relationship between N 1 and volume consumed, even though perceived decision freedom did increase as Nalt increased.

Walton and Berkowitz were also interested in how Nalt and amount available per alternative affect total decision time. The reason behind the interest in amount per alternative is not made clear. Nalt has often been cited as a determinant of decision time, but it is prudent to examine the record.

Factors, Affecting Decision Time

The authors refer to Hendrick, Mills, and Kiesler (1968) as support for the general proposition that decision times first increase, then decrease, as the complexity of the choice increases, with Nalt seen as one factor contributing to complexity. Unfortunately, the authors misconstrue the oft-cited Hendrick et al data, just as Hendrick et al and legions of others have done, a point documented at some length elsewhere (Wright, 1978). Briefly, Hendrick et al contrasted subjects facing (a) four attractive neckties or (b) two attractive and two unattractive neckties when (i) one cue was made salient or (ii) many cues were supposedly salient. Subjects chose twice, in different conditions of the design. On the first choice problem, extra attractive alternatives caused higher total decision times when one cue was salient. When many cues were salient, there was no reliable effect due to extra alternatives. In the second problem, extra attractive alternatives led to higher decision times regardless of how many cues were salient. Hence, in no case did increasing the proportion of attractive alternatives in the set cause increases in decision times. Hendrick et al stated otherwise in their conclusion: "Under the many dimension condition, decision time was shorter for four equally attractive alternatives than for two equally attractive and two unattractive alternatives (1968, p. 317)". The study is so often cited and the design or data so often mischaracterized that it deserved comment. (For more discussion of the data from Hendrick et al, see Wright, 1978).

It also seems timely to take stock of the accumulated evidence regarding Nalt effects on times to decide between complex alternatives. Many researchers have proposed an inverted-U shaped relationship between decision times and a choice problem's complexity, with Nalt or the number of salient cues to be used for evaluations (Ncue) seen as contributors to complexity. Empirical tests include Kiesler, 1966, Pollay, 1970; Jacoby, Speller, and Kohn, 1974; Jacoby, Szybillo, and Busato-Schach, 1977, and Van Raiij, 1976, in addition to Walton and Berkowitz. Often, manipulations of Nalt have been confounded with manipulations of the disparity between alternatives in the set, a problem Walton and Berkowitz apparently avoided. And manipulations of Ncue have likewise been questionable (Wright, 1978). To date, no study has produced clear evidence that increasing either Nalt or Ncue shortens someone's total decision time, except Van Raiij who used a within-subject design (which may have heightened subjects awareness of the Ncue manipulation). Further, even if such an effect had been empirically documented, it would not necessarily be due to increased complexity; it is quite possible to simplify someone's choice problem by adding an extra cue, providing it is a discriminator cue.

Research Strategy

Two final comments are offered dealing with general research strategy where decision times are of interest. First, Walton and Berkowitz predict that adding more flavors to the choice set in their experiment will cause longer decision time, since this whole problem seems intuitively to be relatively simple. With a complex variable like "choice problem complexity" and a hypothetical humpbacked U-shaped relationship in mind, banking on intuition about just where one's manipulation will tap into the "complexity" continuum seems very risky. Depending on the other problem parameters which may affect complexity, an incremental increase in Nalt might push times up, bracket the hypothetical inflection point, or pull times down, and a researcher's intuition about where the inflection point occurs in a given setting may not be good. In any case, given the wonderful flexibility of U-shaped hypotheses in handling data, it seems desirable to offer strong reasoning or supporting empirical results to show that a predicted relationship of form X was justifiable a priori.

Second, it seems desirable to recognize that total decision time is spent performing conceptually different activities, that these activities are what is really of concern, and that a given total time prediction should be reasoned out in terms of the relevant activities. For example, in predicting that there will be no difference in total times between "two drinks, one pitcher" and "two drinks, two pitchers each", the authors apparently assumed that physical scanning time, which should be somewhat longer when there are twice as many pitchers to inspect, would be a trivial component of total decision time. This should at least be explicitly discussed. Further, is the increase in time when Nalt doubled in this experiment due to increased air scanning time or increased preference analysis time or both? Spelling out the thinking about the relevant cognitive activities that may be affected by a problem parameter is desirable, as well as seeking data analysis methods that enable one to identify which activity was affected.


Raju's interest is more in the line of taxonomy development. He examines the intercorrelations between a sampling of information-seeking acts, between a set of general dimensions of new products, and between these two sets of data, relying on reported perceptions about product attributes and about action intentions.

Research Strategy

One aspect of Raju's research strategy that deserves comment is the treatment of the "action" variables. Raju was interested in the types of information seeking acts that people see as similar in Judging their information search intentions. To explore this, the intercorrelations between reported intentions to perform four information-seeking acts, as evoked by numerous stimulus products, were examined. As it turned out, three of the acts (Numbers 2,3 and 5 in Table 2, Raju, 1978) were closely correlated. These three have in common that a person "happens to notice" the actual product, and this stimulates immediate information search. One might hypothesize that these acts represented minor variants of a general behavioral factor we might call "physical-product stimulated search". Note that Behavior #4, which did not correlate closely with the other three, is the only one where search is triggered by a stimulus (an ad) other than the physical product. However, this interpretation may be questioned because Behavior #4 also differed by depicting an information seeking act with a deferred payoff; the data only arrive much later. Hence, it is unclear how to interpret the pattern of intercorrelations.

How might this question of the types of information seeking acts that people see as similar or different in judging their action intent be studied? Two strategies might be used. One might begin by assembling as exhaustive a list as is possible of the specific information seeking acts or action contexts a person might envision. A large random sample of these acts, or even the entire set, might then be used as separate intent measures in a study like Raju's, where products are the stimulus.

An example of this research strategy is the numerous studies by Triandis and his colleagues (Triandis, 1972) regarding the dimensions of "interpersonal action". First, a broad list of concrete interpersonal acts was compiled, using verbs as the basis. Then subjects reacted to stimulus persons described in various ways by stating their intentions to perform the various interpersonal acts on the lists. A factor analysis identified a small set of inter-personal-action contexts to which people seem to react: Self-superior contexts, self-subordinate contexts, intimate act contexts, etc. Replications using the same "start from scratch" methodology in different cultures have yielded evidence of much cross cultural similarity in the dimensions, although different tendencies are noted in the way specific "con-text-other person" combinations are evaluated.

A second approach would be to hypothesize different clusterings of acts, based on prior evidence or some rudimentary theory about what makes one act of information seeking different from another. Then one could pick a few specific acts intended to represent each of the hypothesized clusters, to be used as intent measures.

In picking concrete acts or action contexts, one would consciously strive to clearly differentiate the acts along one's hypothesized dimensions. For example, if one thought that "physical product stimulated search" differs in people's minds from "symbolic product stimulated search", and that "quickly completed acts of search" differ from "wait for the data to arrive" acts, one would create acts which keep these dimensions distinct, for inclusion in the set that elicits intent statements: for example, "notice it in a store, pick it up and look at it" (physical product stimulated/immediate data), "notice it in a vacationing friend's house/leave a note asking for advice about it when the friend returns"(physical product stimulated/ delayed data), "notice an ad for it, read the ad" (symbolic product stimulated/immediate data), "notice an announcement for a brochure, send for the brochure" (symbolic product/delayed data). The author and Alain Cousineau tried this approach several years ago. We identified over 20 concrete acts a person could perform regarding a product, expressed in general terms. These were combinations of four acts (uses, gives as gift, seeks advice about, gives advice about) and five "other people", intended to represent the five types of interpersonal acts identified by Triandis' research: self (private), sweetheart, boss, employee, and acquaintance. About two dozen subjects judged their intent to perform each of the acts regarding stimulus products described by wide variations in four general factors: familiarity, expense, whether or not use or consumption would be publicly conspicuous, and whether or not the product was closely related to the person's self-image. Obviously, our motivation was not too dissimilar from Raju's. We never wrote up the results of the factor analysis performed on the behavioral intent reports, because of the small sample size and because the response task fatigued subjects and this may have influenced their responses somewhat. Our research question was whether the inherent nature of the act (using, giving, seeking information, giving information) or the nature of the people who are part of the act attracts people's attention. For what they are worth, our data showed that the inherent differences between the acts attracted little interest. The basis for discrimination was the degree of intimacy implied by the person(s) involved in the acts.

A second question of research strategy concerns the use of cross-subject correlations rather than with-in-subject correlations. With 20 stimulus products, multiple observations are available for each of Raju's subjects. Even allowing for a reduction due to "product irrelevance", it seems possible to test the covariation between limited sets of the variables on a within-subject basis. This would yield an individual-level test of a basically individual level theory, and could alleviate problems in interpreting cross-sectional correlations caused by individual differences in scale usage. Whether or not one should use within-subject or cross-sectional designs and data analysis methods is a complex issue but since a within-subject design was used for data collection, within-subject data analyses might be productive.


Our concern must be with the theoretical arguments Harvey develops, since an empirical test is not discussed. Harvey defines a construct called "evaluative conflict" in terms of the disparity between the values associated with a product's attributes, then uses Kelman and Baron's (1968) taxonomy of conflict resolution mechanisms to derive the propositions that (i) evaluating a high-risk innovation is conceived as a "single-goal" decision problem in which people are motivated to resolve the disparity between attributes by "confronting" it, and (ii) that this implies extensive search for external evidence from trusted sources.

As a prelude, one might note that Kelman and Baron's original propositions about conditional modes for inconsistency resolution have never been tested empirically, to the author's knowledge. Since the theorizing is interesting, one must assume that the lack of testing stems from difficulties with interpreting or operationalizing what K & B mean by "single goal" or "dual goal" situations or by "denial", "compartmentalization", "transcendence" etc. It is natural therefore to examine how Harvey deals with this translation.

Harvey confines his interest explicitly to high risk innovations, and asserts that the associated decision process is usually a "single-goal" process. This means, using Harvey's wording, that the discrepant elements (i.e., the disparate product characteristics) are tied to the same goal state. In a two goal (or "n"-goal) situation, the disparate product attributes would be linked to different goal states.

The author would restate what K & B had in mind this way: Assume there are two dimensions, " and B, being used for the evaluation. " is related to one type of goal or consequence (e.g., personal health), and B to a different type of consequence (e.g., aesthetic pleasure). One type of conflict that can arise is if one receives disparate evidence about the product's ability to cause "aesthetic pleasure"; some evidence suggests that it will and some that it won't. This may be termed "within-dimension uncertainty". Another type of conflict arises if one thinks the product is good on one dimension and bad on the other. One knows with great certainty that the product has a good and a bad side to it. We can call this "cross-attribute disparity". [These terms evolved from discussions earlier discussions with Fredrick Winter.] "Within-dimension uncertainty" corresponds, in my thinking, to K & B's "single goal" conflict; "cross-attribute disparity" corresponds to a "multi-goal" conflict.

If the above discussion, captures what K and B meant in terms of the structure of one's beliefs about a product's attributes, then Harvey's "evaluative conflict" construct seems to be defined as cross-attribute disparity. If so, his domain is by definition a two-goal ( or n-goal) conflict situation, and this would lead him (at least, using K and B taxonomies) to view "maintenance/confrontation" strategies, not "reduction/confrontation" strategies as the preferred mechanisms for handling evaluative conflicts stirred by risky innovations. More discussion about the link between Harvey's evaluative conflict notion and K and B's single-and dual-goal notions seems desirable or, more directly, about the way people deal with cross-attribute disparity vs. within-dimension uncertainty .

In highlighting external information search as a preferred strategy for conflict resolution, Harvey actually introduces a quite different resolution strategy into K and B's scheme. K and B's 12 modes all are intended to represent strategies of self-initiated cognitive adjustment (rethinking the problem), unaided by external data. The sole exception seems to be an "action change", but this does not require now data. If, as Harvey argues, a single-goal, central value situation is of interest, K and B propose as "reduction/confrontation" modes a change in one's global evaluation of the product, a change in one's own action toward the product, a change in one's norm, or an attempt to persuade one's source of information that he/she is wrong, or to change the product itself. None of these involves a search for more information. Nor do any of the strategies K and B suggest for dual-goal, central value situations (bolstering, differentiation, transcendence) involve a search for now external data.

So Harvey in fact uses K and B's structure up to a point, then reasons beyond the internal modes of resolution they deal with to modes involving external search. This seems like a potentially productive line of reasoning, but it requires more development. For example, many of the mechanisms K and B describe could, perhaps, include some external information search (source derogation, compartmentalization, attitude change, bolstering, for example), but these appear in every quadrant of their taxonomy. It seems useful to reason in as much detail about information search strategies as K and B did about cognitive readjustment Strategies. A proposition that more external information will be sought from trustworthy sources seems too general. What types of information seeking activities do the respective situations K and B describe elicit? With what purpose, specifically? Why would one seek more data if one experienced cross-attribute disparity? To find another dimension to act as a tie-breaker? To make oneself more uncertain about one attribute? What if one's calculation of cross-attribute disparity already encompasses the major evaluative dimensions?

In summary, it seems quite useful to integrate theorizing about conditional information search strategies with K and B's interesting theorizing about self-initiated cognitive adjustment strategies, and Harvey points us in that direction. More detailed reasoning regarding the nuances of information search activities, perhaps in collaboration with Raju, should prove very enlightening. Then, the nontrivial problems of operationally distinguishing the situations and especially the varied types of information search activities can be confronted.


Harvey, J. W., "Evaluative Conflict and Information Search in the Adoption Process", paper presented at the Association of Consumer Research Conference, Miami, 1978.

Hendrick, C., Mills, J., and Kiesler, C. S., "Decision Time as a Function of the Number and Complexity of Equally Attractive Alternatives", Journal of Personality and Social Psychology, 8 (1968), 313-318.

Jacoby, J., Speller, D., and Kohn C., "Brand Choice as a Function of Information Load: Replication and Extension", Journal of Consumer Research, 1 (1974), 33-42.

Jacoby, J., Szybillo G., and Busato-Schach, "Information Acquisition in Brand Choice Situations", Journal of Consumer Research, 3 (1977), 209-216.

Kelman, H. C. and Baron, R. M., "Determinants of Modes of Resolving Inconsistency Dilemmas: A Functional Analysis", in R. Abelson et al (Eds.), Theories of Cognitive Consistence: A Sourcebook. Chicago: Rand McNally, 1968.

Kiesler, C. A., "Conflict and the Number of Choice Alternatives", Psychological Reports, 18 (1966), 603-610.

Raju, P.S., "Stimulus-Response Variables in New Product Research", paper presented at the Association for Consumer Research Conference, Miami, 1978

Reibstein, D. J., Youngblood, S. A. and Fromkin, H. L., "Number of Choices and Perceived Decision Freedom as a Determinant of Satisfaction and Consumer Behavior", Journal of Applied Psychology, 60 (1975), 434-437.

Triandis, H. C., The Analysis of Subjective Culture. New York: Wiley, 1972.

Van Raiij, F., Consumer Choice Behavior: An Information Processing Approach. Ph.D. Dissertation, Tilburg (Hol.) University, 1977.

Walton, J. R. and Berkowitz, E. N., "The Effects of Choice Complexity and Decision Freedom on Consumer Choice Behavior", paper presented at the Association for Consumer Research Conference, Miami, 1978.

Wright, P., "Decision Times and Processes on Complex Decision Problems", Faculty Working Paper, Graduate School of Business, Stanford University, 1978.