Predicting Behavior With Intentions: a Comparison of Conditional Versus Direct Measures

ABSTRACT - This research attempts to extend upon an earlier study (Warshaw, 1980) concerning the predictive validity of alternative behavioral intentions measures. Contrary to Warshaw's findings, our results indicate little difference between conditional and direct measures. In addition, contextual correspondence between intention and behavior had little effect on their relationship.

Citation:

Paul W. Miniard, Carl Obermiller, and Thomas J. Page, Jr. (1982) ,"Predicting Behavior With Intentions: a Comparison of Conditional Versus Direct Measures", in NA - Advances in Consumer Research Volume 09, eds. Andrew Mitchell, Ann Abor, MI : Association for Consumer Research, Pages: 461-464.

Advances in Consumer Research Volume 9, 1982      Pages 461-464

PREDICTING BEHAVIOR WITH INTENTIONS: A COMPARISON OF CONDITIONAL VERSUS DIRECT MEASURES

Paul W. Miniard, The Ohio State University

Carl Obermiller (student), The Ohio State University

Thomas J. Page, Jr. (student), The Ohio State University

ABSTRACT -

This research attempts to extend upon an earlier study (Warshaw, 1980) concerning the predictive validity of alternative behavioral intentions measures. Contrary to Warshaw's findings, our results indicate little difference between conditional and direct measures. In addition, contextual correspondence between intention and behavior had little effect on their relationship.

INTRODUCTION

Behavioral intention has long been an important construct in consumer research. Theoretical treatments (e.g., Engel, Kollat, and Blackwell, 1978; Howard and Sheth, 1969) position behavioral intention as intervening between attitude and behavior. Empirical applications have demonstrated that behavioral intention can be used to predict behavior (e.g., Stapel, 1968; Wilson, Mathews, and Harvey, 1975; Ryan and Bonfield, 1980). Given its importance both as a theoretical construct and as a useful predictor variable, questions regarding the appropriate measurement of behavioral intentions are critical to research in the area. This is particularly true since the large body of published research employing paper and pencil measures of intention is small in comparison with the number of real-world implications in consumer surveys.

Despite the importance given behavioral intention in consumer behavior theory and the widespread use of intention measures in practice, many researchers have found low correlations between measured intentions and observed behavior. One area of explanation for such predictive variations has focused on the manner in which intention is assessed. Ajzen and Fishbein (1980) claim that the strength of the intention-behavior relationship will be influenced by the degree of correspondence between the intention measure and the behavior criterion. Prediction should be enhanced when the measure of intention and behavior correspond in terms of target, action, context, and time elements. Intentions to purchase an automobile (a general target), for example, should not be expected to accurately predict purchases of a particular automobile brand (a specific target). Similarly, behaviors that can be performed in a number of contexts or situations may not be accurately predicted when the criterion behavior is to be observed in each of these contexts and the intention measure fails to fully specify the situations in which the behavior can occur.

Evidence supporting the predictive need for correspondence between predictor and criterion variables has been generated by research investigating attitude-behavior relationships. In an extensive review of this literature, Ajzen and Fishbein (1977) found that strong relationships between attitudinal and behavioral measures existed under high levels of correspondence whereas low levels generated weak relationships. They noted, however, that there is a conspicuous lack of research relevant to the need for contextual correspondence. More recently, Miller and Ginter (1979) examined the value of employing contextually correspondent attitude measures for predicting patronage behavior of fast-food outlets. Their research indicated that a contextually correspondent attitude measure provided a superior prediction of patronage behavior than attained by an attitude measure lacking contextual correspondence.

Warshaw (1980) has also argued that the lack of contextually specific measures may partially account for low intention-behavior relationships, and has recommended measuring the contextual antecedents of intentions. If these conditional antecedents are specified completely and independently, then intentions can be decomposed as follows:

EQUATION

where BIy is the likelihood of performing behavior y, P(Xi) is the likelihood of conditional antecedent Xi, P(Y/Xi) is the likelihood of performing behavior y given conditional antecedent Xi, and" is the number of relevant conditional states.

One possible contextual antecedent of product purchase is purchase location. Soft drinks, for example, can be bought in a number of acquisition sites (e.g., store, vending machine, restaurant). For estimating intention to purchase a given soft drink brand, assessment of P(Xi) would involve asking respondents the likelihood of produce purchase from each location (e.g., How likely is it that you will buy a soft drink from a vending machine?). Measures of P(Y/Xi) would require respondents to estimate the probability of purchasing a particular brand of soft drink given that product purchase occurred at each acquisition site (e.g., How likely is it that you would buy a Coke if you purchased ; a soft drink from a vending machine?).

Warshaw (1980) compared conditional and direct measures of intentions to purchase various soft drinks and concluded that a conditional measure employing purchase location as the conditional antecedent was superior in predicting purchase behavior. It is not clear, however, whether these differences were due to the measurement format or the lack of equivalence in contextual correspondence as both factors covaried together. That is, the direct measure of purchase intention did not specify the acquisition sites that were incorporated into the behavioral measure and thus lacked contextual correspondence with the criterion behavior. In contrast, because purchase location represented the antecedent upon which the conditional measure was based, this measure did possess contextual correspondence. One cannot, then, unambiguously determine which of these two measurement issues contributed to Warshaw's results. In a comparison of direct and conditional measures that were equivalent in their contextual correspondence, Jaccard, Knox, and Brinberg (1979) found the two approaches to yield essentially equivalent predictions of behavior. This would imply that the differences reported by Warshaw were due to the inequities in the measures' contextual correspondence. The goal of this research was to disentangle these factors by e a mining conditional and direct measures that contextually corresponded to the criterion behavior.

METHOD

Subjects and Procedures

Subjects were 66 junior and senior level undergraduates enrolled-in various marketing courses. On Wednesday morning subjects received a questionnaire containing the direct and conditional measures;of likelihood of purchasing seven brands of soft drink between that moment and the-following Monday morning. The brands were Coke, Pepsi, 7-Up, Dr. Pepper, Sprite, Tab, and "other brand. " These brands were identified in a pretest done several weeks earlier as the most popular with the subjects. The following Monday afternoon, subjects reported purchases for each brand in three purchase locations (restaurant/ bar, store, and vending machine).

Questionnaires

Subjects were randomly assigned one of two questionnaire versions. Thirty-two subjects received questionnaires with noncontextual direct measures of intention. The noncontextual direct items read "What is the probability _ had you will purchase 'given brand' between now and next Monday?" The remaining 34 subjects received a questionnaire version with a modified direct measure of intention, which read "What is the probability that you will purchase 'given brand' from a store, vending machine, or at a restaurant/bar between now and next Monday?" The modification in the second version of the questionnaire was designed to equalize the contextual correspondence of the direct and conditional measures.

Both versions of the questionnaire included items required for the conditional measure of intention suggested by Warshaw (1980). These items were:

"What is the probability that you will make one or more purchases of soft drink from 'given location' between now and next-Monday?" (WL)

"Assuming you do make one or more purchases of soft drink from 'given location' between now and next Monday, what is the probability that you will, in total, buy only one brand of soft drink rather than more than one brand?" (XL1)

The conditional probability of buying more than one brand (XL2) from one location was derived by 1-XL1.

"Assuming you do purchase one and only one brand of soft drink from 'given location' between now and next Monday, what is the probability that it will be (each of seven brands listed)?" (ZLli)

"Assuming you do purchase more than one brand of soft drink from 'given location' between now and next Monday morning, what is the probability that each brand below will be one of those two or more brands you select (each brand listed)?" (ZL2i)

All items were scored on an 11-point percentage scale ranging from 0 tn 100%.

According to Warshaw (1980) the conditional probability of purchase is derived from the preceding measures in the following manner:

1) The raw scale values (ZLli) were standardized:

EQUATION

This guaranteed that for each subject EQUATION.

2) For each location (L), intentions were measured as BI_Li = W_L (X_L1 x Z_Lli + X_L2 x Z_L2i).

3) The overall derived BI for each brand was one minus the joint probability of not buying from any location, where

EQUATION

is the probability of not buying brand i, and the derived purchase intention is given by  EQUATION.

RESULTS AND DISCUSSION

There are several ways to evaluate the performance of intention measures. Warshaw (1980) divided subjects into purchasers and nonpurchasers of each brand. He asserted that the ideal values of a behavioral intention measure are 1.0 for purchasers and 0.0 for nonpurchasers. His analysis, then, consisted of paired-comparison t-tests on the mean purchase probabilities for the conditional and direct measures. Table 1 presents the data for this analysis from the present study.

The upper half of Table 1 shows the results for the questionnaire condition involving the noncontextual direct measure. Purchasers' responses to the direct and conditional measures statistically (p < .03) differed (in favor of the direct measure) on only the "other" brand. Across brands, the direct measure (X=.634) produced a slightly higher estimate than the conditional format (X=.599), although this difference was not statistically significant (p > .2). For nonpurchasers, the conditional measure yielded lower (p < .01) estimates overall (.185 vs. .223). Only this latter finding is consistent with Warshaw's results. The absolute level of intention estimates for nonpurchasers are higher in this study; the absolute levels for purchasers are about the same. (Warshaw's results:

Purchasers: BI_direct - .539, BI_conditional = .652;

Nonpurchasers: BI_direct = .077, BI_conditional = .051).

The lower half of Table 1 contains the results for the contextually correspondent direct measure versus the conditional measure. For Coke (p < .03) and Tab (p < .08), the direct measure gave statistically significantly higher estimates, but overall the difference between the direct (X=.620) and conditional (X=.607) measures was not statistically significant (p > .59). For nonpurchasers, the conditional format (X=.218) again produced lower (p < .01) estimates than the direct measure (X=.258).

Using Warshaw's analytical approach, then, the present results replicated Warshaw's findings for nonpurchasers but failed to reproduce his results for purchasers. In addition, changing the level of contextual correspondence of the direct measure did not affect its performance relative to the conditional measure. It is interesting to note that purchasers indicated a greater (p < .01) probability of purchase than did non purchasers for both the direct and conditional measures.

TABLE 1

COMPARISON OF DIRECT AND CONDITIONAL MEASURES

The soundness of the preceding analysis is directly dependent upon the validity of assuming that the ideal estimates for purchasers and nonpurchasers are 1.0 and 0.0 respectively. This conception confuses behavioral intention, a continuous variable, with behavior, a dichotomous variable. One may well grant that the true mean behavioral intention of purchasers is likely to be higher than that of nonpurchasers, but there are certainly many instances in which behavioral performance (nonperformance) is not preceded by an intention of 1.0 (0.0). Moreover, for predictive purposes, it is unnecessary for stated intentions to correspond with measures of purchase so long as they correlate with them. Thus, measures that produce means of .7 and .3 for purchasers and nonpurchasers respectively will classify behavior as well as measures that produce means of 1.0 and 0.0. Therefore, both between and within-subject correlational procedures were employed for examining the measures' predictive power. The former method involves correlating intentions with purchase within each brand across subjects. Since behavior is a dichotomous variable (i.e.> purchaser either did or did not occur), a point-biserial correlational test was used. The results (Table 2) of the between-subjects correlational analysis indicate little difference in predictive ability. For the comparison of the noncontextual direct measure with the conditional format measure, the latter predicted purchase of Pepsi better and Sprite worse (p < .05). For the comparison of the contextual direct measure with the conditional measure, no differences were statistically significant (p > .1).

For the within-subjects analysis, a point-biserial correlation was computed between intentions and purchase behavior across brands for each subject. This approach may be superior to the former method (i.e., between-subjects) since it avoids predictive attenuations arising from scale interpretation differences among respondents. Each individual's correlation was converted using an r to z transformation and these z scores were employed for estimating the measures' average correlation and comparing their predictive power. Three subjects were eliminated from this analysis since they failed to purchase any soft drink during the specified time interval.

TABLE 2

BETWEEN-SUBJECTS CORRELATIONS BETWEEN INTENTION MEASURES AND BEHAVIOR

The results of this analysis revealed little predictive difference between the direct and conditional measures. In the condition involving the noncontextual direct measure, direct (r=.72) and conditional (r=.76) measures provided very similar (p > .59) predictions. Similarly, the two measures did not differ in their predictive power (p > .5) when the direct measure contextually corresponded to the criterion behavior. In this condition, the intention-behavior correlation was .71 for the direct measure and .67 for the conditional measure.

It should be noted that the present correlations between intentions and behavior are somewhat higher than those obtained in many marketing investigations. This may be due to the nature of the product category and/or the short time interval between the intention and behavior measures. Alternatively, the size of the correlations may be due to bias stemming from reliance upon self-reported behavior.

Finally, the preceding analyses have suggested that the direct measure's predictive power was unaffected by its contextual correspondence with behavior. To specifically examine this issue, the intention-behavior correlations involving the contextual and noncontextual direct intention measures were compared. Since existing evidence suggests that increases in contextual correspondence should enhance prediction, a one-tailed test was employed. For the between-subjects correlations (Table 2), the only noteworthy predictive difference occurred for Coke where the contextual direct measure tended to provide a superior prediction (p < .07). For the within-subjects correlations, the contextual (r=.71) and noncontextual (r=.72) measures yielded virtually identical predictions of purchase behavior.

CONCLUSION

The two aims of this research were to (1) compare the predictive power of conditional and direct measures of intention and (2) examine the need for contextual correspondence between intention and behavior measures. In attempting to replicate Warshaw's study involving the first research coal. we failed to fully duplicate his results. When we used the same method of analysis, we found the conditional measure "better" for nonpurchasers but not for purchasers. No explanation is readily available. Several differences existed between the present and Warshaw's study: subjects were American students rather than Canadian housewives, data were gathered by questionnaire rather than personal interview, and "other location" (included in the Warshaw study but dropped in his analysis) was not employed in the present investigation as one of the contextual antecedents. It is not clear, however, how these differences could account for the discrepancies in results between the two studies.

Because of the previously discussed problems we have with Warshaw's mode of analysis, we would place more weight on the correlational results for evaluating the measure's relative merits. These analyses suggested that the two measurement approaches were equivalent in their predictive power. Since the conditional format required nine times as many measurements in this study, the present evidence does not justify its use for predictive purposes.

Ajzen and Fishbein (1980) have argued that the strength of the intention-behavior relationship will be influenced by the degree of correspondence between the intention measure and the behavioral criterion. For the present investigation, there would seem two potential conditions under which the contextual measure could outperform the noncontextual measure. First, to the extent specification of purchase locations reminds the person of a location which otherwise might not have been considered and in which purchase occurs, contextual correspondence should enhance prediction. Second, if the person considers some location other than those employed in the behavior measure (i.e., other than store, vending machine, and restaurant/bar in the present study) in responding to the noncontextual measure, then this could lower the predictive power of the noncontextual measure relative to the contextual measure. Surprisingly, contextual correspondence had little effect upon the direct intention measure's predictive accuracy, thus suggesting that neither of the above conditions were operative in this particular test situation. Further research concerning the role of contextual correspondence in the intention-behavior relationship would seem desirable.

REFERENCES

Ajzen, I. and Fishbein, M. (Sept., 1977), "Attitude-Behavior Relations: A Theoretical Analysis and Review of Empirical Research," Psychological Bulletin, 84, 888-918.

Ajzen, I. and Fishbein, M. (1980), Understanding Attitudes and Predicting Social Behavior (Englewood Cliffs, NJ: Prentice-Hall).

Engel, J., Blackwell, R. and Kollat, D. (1978), Consumer Behavior, 3rd Ed., (Hinsdale, IL: Dryden Press).

Howard, J. and Sheth, J. (1969), The Theory of Buyer Behavior, (New York: John Wiley and Sons, Inc.).

Jaccard, J., Knox, R. and Brinberg, D. (1979), "Prediction of Behavior from Beliefs: An Extension and Test of a Subjective Probability Model," Journal of Personality and Social Psychology, 37, 1239-1248.

Miller, R. E. and Ginter, J. L. (Feb., 1979), "An Investigation of Situational Variations in Brand Choice Behavior and Attitude," Journal of Marketing Research, 16 9 111-123.

Ryan, J. and Bonfield, E. H, (Spring, 1980), "Fishbein's a Extended Model: A Test of External and Pragmatic Validity," Journal of Marketing, 44, 82-95.

Stapel, J. (1968), Predictive Attitudes," Attitude Research on the Rocks, L. Adler and I. Crespi, eds., (Chicago: American Marketing Association).

Warshaw, R. (February, 1980), "Predicting Purchase and Other Behaviors from General and Contextually Specific Intentions," Journal of Marketing Research, 17, 26-33.

Wilson, T., Mathews, H. Lee and Harvey, James W. (March, 1975), "An Empirical Test of the Fishbein Behavioral Intentions Model," Journal of Consumer Research, 1, 39-48.

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