The Impact of Information Load and Variability on Choice Accuracy

Roger J. Best, University of Oregon
Michael Ursic, Gonzaga University
ABSTRACT - Past studies have concentrated on examining the impact of the number of brands and attributes on decision accuracy. Yet the amount of information is only one dimension of the complexity of the task. This research found that the alternative variance and the similarity of the choices has more influence on decision accuracy than numbers of brands or attributes.
[ to cite ]:
Roger J. Best and Michael Ursic (1987) ,"The Impact of Information Load and Variability on Choice Accuracy", in NA - Advances in Consumer Research Volume 14, eds. Melanie Wallendorf and Paul Anderson, Provo, UT : Association for Consumer Research, Pages: 106-108.

Advances in Consumer Research Volume 14, 1987      Pages 106-108

THE IMPACT OF INFORMATION LOAD AND VARIABILITY ON CHOICE ACCURACY

Roger J. Best, University of Oregon

Michael Ursic, Gonzaga University

ABSTRACT -

Past studies have concentrated on examining the impact of the number of brands and attributes on decision accuracy. Yet the amount of information is only one dimension of the complexity of the task. This research found that the alternative variance and the similarity of the choices has more influence on decision accuracy than numbers of brands or attributes.

INTRODUCTION

Despite the number of studies on the relationship between the amount of information and decision accuracy there are still no conclusive results. Studies by Jacoby and his associates (1974a; 1974b) and Malhotra (1982a) characterized decision accuracy as increasing with more information to a point and then decreasing when additional information is given. Several other papers suggested that choice accuracy increases with more information, (Russo 1974; Wilkie 1974; Summer 1974; Malhotra 1982b). Others have shown that the amount of information does not influence decision accuracy (Oskamp 1965; Slovic, 1971). Clearly, more research is needed to clarify the nature of these relationships.

Also, there is a lack of studies on the relationship between other dimensions of task complexity and decision accuracy. Past studies have concentrated on the number of brands and attributes, yet these are only two elements of information complexity. Two other possible dimensions are choice similarity and alternative variance.

Alternative variance is the extent to which all the pieces of information about a brand is in agreement. For instance if a choice was poor on one attribute, very good on another and fair on a third attribute, then there would be high variance. Supposedly, it would be more difficult for a consumer to process information that is highly variable.

The similarity of the choice! is the extent to which all choices are equal in desirability. It might be more difficult for a consumer to choose the most appropriate brand when all the available brands are roughly equal in desirability than it is when individual brands are clearly superior to others. Clearly work is needed to examine the relationship between variance and choice similarity on decision accuracy.

The purpose of this paper then is to examine the impact of the number of brands, number of attributes, alternative variance and choice similarity on decision accuracy. Hopefully, this research will yield additional insight concerning the relative relationship between various complexity measures and information overload.

METHOD

Subjects and Procedures

One hundred-sixty university students participated in the study. The median age was twenty-one and sixty-one percent were male. Because students were used, products selected for study reflected the interest and experiences of this consumer segment. For this reason ten-speed bicycles, hair conditioners, and rental apartments were selected for evaluation.

The procedures used in this study were as follows. First, eight important attributes for each product class were selected from a pretest of thirty students and from an examination of Consumer Reports. Next, the one hundred and sixty subjects were asked to rank-order the importance of these attributes. Information boards were then created for each product class, like that shown in Figure 1, where the subjects were given several hypothetical brands. These brands were described from very good to very poor on the subjects' most important attributes.

The order of the attributes on the page was the same as the subjects' importance rankings. Subjects were then asked to choose their most preferred brand and then reevaluate the importance of the attributes on a 100-point constant-sum scale.

FIGURE 1

SAMPLE CHOICE SITUATIONS

Experimental Treatments

As shown in Table 1 a 4x2x2 factorial design was used to operationalize the treatments in which:

Attributes used to describe alternatives varied over four levels (2,4,6, and 8 characteristics).

Number of brands varied across two levels (3 and 6 brands).

Choice similarity was varied across two levels (no difference between al and discernible difference between alternatives).

Alternative variance for a given brand was varied over a range from very high variability of the information to very low variability.

TABLE 1

EXPERIMENTAL DESIGN

Choice similarity and alternative variance were quantified as follows. In the condition where the choices were similar the unweighted score of each alternative, computed by assigning '5' to very good, '4' to good, '3' to fair, '2' to poor, and '1' to very poor, was nearly equal. In the dissimilar condition there was a distinct hierarchy of unweighted scores. Alternative variance was computed for each brand using the above coding scheme and then an average was taken for the entire brand set.

Note that descriptive evaluators (very good, good, etc.) were used instead of raw data (49 cents, 96 cents, etc.). This procedure has three advantages. It reduces interindividual differences in perception. It allows for precise and standardized computation of the choice similarity and alternative variance conditions. It removes the necessity for using ideal points in the computation of decision accuracy. This is because each individual can interpret a descriptor in a manner consistent with his or her preference.

Choice Accuracy

A subject's best alternative was determined by multiplying the subjects' importance weight from the constant sum scales for each of the given attributes by the performance of the alternative on each attribute, and then summing the results. When the actual choice matched the best choice, an "accurate" decision was made. The chosen brand, rather than a rank-order of preference, was used to make the process more like a purchase decision.

While there are many ways to quantify decision accuracy, the procedure used in this research is appropriate for several reasons. First, it is consistent with past research in consumer-oriented journals (Jacoby et al 1974a; Malhotra 1982a). Second, it is preferable to other suggested measures of decision accuracy, such as expert opinion. Experts don't always agree and many experts feel that there is no objective way to determine decision accuracy. Other suggested measures also have severe problems (Jacoby 1977).

Analysis

As explained the dependent variable was a binary depending on whether an accurate choice was made, which means that LOGIT is an appropriate analysis technique. There is much support for the use of LOGIT when analyzing this type of categorical variable. (Bishop, et al 1975; Malhotra 1982a: 1982b).

Using the LOGIT framework the experimental treatments created in this study can be represented as follows:

             ( Pi )

loge                      = X1 + X2B2 + X3B3 + X4B4 + X5B5

            ( 1-Pi )

where:

Pi = Proportion of correct choices for a given choice situation.

X1 = Intercept term.

B2 = Number of characteristics (2, 4, 6, or 8).

B3 = Number of alternatives (3 or 6).

B4 = Choice similarity (distinct hierarchy of choices, all choices similar).

B5= Alternative variance (continuous choices from very small to very large differences).

X2,X3,X4,X5 = Importance of these treatments to choice accuracy.

To adjust the probability of a correct choice for chance, the following modification has been recommended (Fleiss 1975) and was incorporated in this study:

            Pio - Pic

Pi =                          

              1 - Pic

where:

Pi = probability of correct choice in experimental condition i adjusted for chance selection of an alternative.

Pio = the observed probability of correct choice in experimental condition i.

Pic = Probability of a correct choice in experimental condition i by chance alone (1 divided by the number of alternatives).

For each of the sets of relationships all possible two-way interactions were specified to allow for examination of potentially important interactive effects between experimental treatments. The linearity of relationships was examined using relevant plots of the data to further determine modifications to the LOGIT model specification.

RESULTS

Table 2 presents the results of LOGIT analysis for each of the three consumer choice situations. The adjusted R2s were .81, . 58, and .73 respectively for ten-speed bicycles f rental apartments, and hair conditioners models. In all cases the overall significance was very high. Since no interactions were found to be significant and no nonlinear relationships were found, linear, additive LOGIT models were used to obtain a very good statistical fit between choice accuracy and the four experimental treatments considered in this study.

Attributes

In all three cases, the addition of more attributes to describe choice alternatives did not improve choice accuracy. As shown in Table 2, these effects were not significant. These results are consistent with previous results reported by (Oskamp 1965; Slovic 1975). Thus, the assumption that more information should contribute to improved choice accuracy is not supported by this research.

Number of Brands

In two cases, ten-speed bicycles and rental apartments, the number of alternatives considered had no significant impact on choice accuracy.

However, for hair conditioners the effect of more brands significantly reduced choice accuracy. It is interesting that only with hair conditioners, the least complex product, did the number of brands affect choice accuracy. This might indicate that the number of brands will only overload consumers in low-involvement situations.

Choice Similarity

When the alternatives were similar, decision accuracy significantly decreased for all three products and this suggests that the similarity of the choices might be an extremely important determinant of decision accuracy that has been ignored by past research. Note that the relationship is strongest in hair conditioners, a simpler product. This might be because people spent less time on the decision and thus the similar choices were more difficult to distinguish.

Alternative Variance

The larger the difference in performance levels used to describe the alternative, the lower the observed choice accuracy. As shown in Table 2, this finding was significant for two of the three product choice situations examined in this study. Ten-speed bicycles and rental apartments were the products where the overload occurred. This might have resulted because people were spending more time processing the information on these products, and thus the increased variability of information created more confusion.

TABLE 2

RELATIVE IMPACT AND SIGNIFICANCE OF EXPERIMENTAL TREATMENTS

DISCUSSION

In two choice situations both alternative variance and choice similarity had a greater impact on decision accuracy than did number of brands and attributes. In the third choice situation choice similarity had a greater impact than number of alternatives and product features. Thus, the results of this research confirm the notion that number of brands and attributes have a limited impact, and that variability factors have a great impact on decision accuracy. Clearly, more research is needed to support or disprove this finding.

More work is also needed to understand the theory behind these relationships. The question of why some dimensions are or are not related to choice accuracy might successfully be addressed by protocol analysis. Further, future studies might incorporate product involvement or complexity as an independent variable. Since complexity of the product seems to influence the results in this research, it is possible that this might be an important construct.

REFERENCES

Bishop, N. M.; Feinberg, S. E.i Holland, P. W. (1975), Discrete Multivariate Analysis: Theory and Practices, Cambridge, MA, MIT Press.

Fleiss, F. L. (1975), "Measuring Agreement Between Two Judges on the Presence or Absence of a Trait," Biometrics, 31. 651-9.

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

Jacoby, J.; Speller, D., and Kohn, C. (1974b), "Brand Choice Behavior as a Function of Information Load," Journal of Market Research, 11, 63-69.

Jacoby, J. (1977), "Information Load and Decision Quality: Some Contested Issues," Journal of Marketing Research, 14, 569-573.

Malhotra, N. (1982a), "Information Load and Consumer Decision Making," Journal of Consumer Research, 8, 419-31.

Malhotra, N.; Jain, A., and Lajakos, S. (1982b), "The Information Overload Controversy: An Alternative Viewpoint," Journal of Marketing, 46, 27-3

Oskamp, S. (1965), "Overconfidences in Case Study Judgments," Journal of Consulting Psychology, 29, 261-265.

Russo, J. (1974), "More Information is Better: A Reevaluation of Jacoby, Sepiler, and Kohn," Journal of Consumer Research, 1, 68-72.

Scammon, D. (1978), "Information Load and Its Effects on the Consumer," Journal of Consumer Research, 3, 148-155.

Slovic, P., and Lichtenstein, S. (1971), "A Comparison of Bayesian and Regression Approaches to the Study of Information Processing in Judgement," Organizational Behavior and Human Performance, 6, 649-744.

Slovic, P. (1975), "Behavioral Patterns of Adhering to a Decision Policy," paper presented at the Institute for Quantitative Research in Finance, Napa, California.

Summers, J. (1974), "Less Information is Better," Journal of Marketing Research," 11, 467-468.

Wilkie, W. (1974), "Analysis of Effects of Information Load," Journal of Marketing Research, 11, 462-466.

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