Uses of Response Certainty in Attitude Measurement

John H. Antil, University of Delaware
ABSTRACT - Accurate measures of attitude are critical if a researcher hopes to obtain high correlations between attitude and behavior. The use of response certainty is shown to be a valuable method to increase attitude-behavior correlations and assist the researcher in interpreting results from attitude measurement. Empirical evidence and theoretical support for the use of response certainty is provided.
[ to cite ]:
John H. Antil (1983) ,"Uses of Response Certainty in Attitude Measurement", in NA - Advances in Consumer Research Volume 10, eds. Richard P. Bagozzi and Alice M. Tybout, Ann Abor, MI : Association for Consumer Research, Pages: 409-415.

Advances in Consumer Research Volume 10, 1983      Pages 409-415

USES OF RESPONSE CERTAINTY IN ATTITUDE MEASUREMENT

John H. Antil, University of Delaware

[The author wishes to express his appreciation to The College of Business and Economics at the University of Delaware and to The Center for the Study of Environmental Policy at Penn State University for financial and other assistance which made this research possible.]

ABSTRACT -

Accurate measures of attitude are critical if a researcher hopes to obtain high correlations between attitude and behavior. The use of response certainty is shown to be a valuable method to increase attitude-behavior correlations and assist the researcher in interpreting results from attitude measurement. Empirical evidence and theoretical support for the use of response certainty is provided.

INTRODUCTION

For many years a primary interest of behavioral scientists has been to predict behavior based upon a measure of attitude. Within the study of consumer behavior, a popular purpose for attitude measurement has been the measurement and prediction of consumers' brand choice (cf. Day 1970; Ryan and Bonfield 1975; Bettman 1979). The and truth, however, is that in spite of widespread theoretical belief in the attitude-behavior relationship, attitudes have not proven to be particularly good predictors of behavior (cf. Wicker 1969 and Wilkie and Pessemier 1973). Though considerable effort has been allocated to explaining the disappointing correlations between attitudes and behavior, new techniques or methodologies that would improve the relationship have been very limited. The popularity within consumer behavior of the Fishbein model (and expectancy value models, in general) from the late sixties to middle seventies is a case in point of a "new" method that led to at best very limited advances in obtaining better predictions of behavior (cf. Wilkie and Pessemier 1973; Fishbein and Ajzen 1975).

This situation should be discouraging to those interested in having not only a reliable and valid method of measuring attitudes but also one that is a good predictor of behavior. Consider this: if a researcher wants to select the best method available (one likely to receive the least criticism and be reasonable to develop) to measure people's attitude toward a concept, what method is he/she likely to select? If one judges by published research and current textbooks, it is very likely that Likert scaling (or perhaps semantic differential) would be selected. What is disturbing about this is that both of these methods have been available for over thirty years. One could conclude from this that in spite of the low correlations obtained between measured attitude and behavior, the methodology in this critical area has not substantially improved for over thirty years.

While certainty is not a new concept, it has received little attention and support in the literature (this is especially true in consumer behavior). The purpose of this paper is to suggest the benefits that can accrue from including response certainty in attitude measurement. It will be demonstrated that the use of this technique can substantially improve correlations between measured attitude and behavior. Further, it will be shown that certainty can be used to evaluate the measurement device and assist in the interpretation of the results. This paper hopes to create new interest and awareness of a method that seems to hold considerable promise for improving attitude measurement.

THEORETICAL BACKGROUND AND LITERATURE REVIEW

The construct of central interest in this paper is certainty or confidence (both terms have similar meaning and are often used interchangeably in the attitude literature). This construct is defined as the degree of certainty or confidence a respondent has in his/her judgments about the attitude object. By itself, the certainty concept has received very little theoretical development and individual attention. Rather, the concept is addressed in the literature as a component of or issue related to other theoretical areas. The following is a brief discussion of four of these areas that are either directly or indirectly related to response certainty.

When one measures the attitudes of several people toward an attitude object it is highly probably that the state or degree of development of the attitudes will differ across individuals (Converse 1970; Day 1970). In other words, one should expect that some respondents are unfamiliar, less knowledgeable or for some other reason(s) have not formed clearly defined attitudes. Further, some individuals may not have formed any attitude at all (i.e. "non-attitudes"). Unfortunately, those individuals with poorly formed or "non-attitudes" will still, when requested, evaluate the attitude object in the same manner as those with well-developed attitude structures. Clearly then, the inclusion of those with poorly developed attitudes will underestimate the degree of the attitude-behavior relationship. Since respondents with poorly developed attitudes should be less certain of their stated positions (Converse 1970; Warland and Sample 1973), measurement of response certainty should be able to identify these individuals and thus allow the researcher to exclude them from the analysis or in some other way adjust for their presence.

Closely related to the concept of poorly developed attitudes is the research from social psychology involving the concept of ego-involvement (cf. Sherif, Sherif and Nebergall 1965) and that from consumer psychology involving low and high involvement (cf. Krugman 1965; Ray 1973; Houston and Rothschild 1978; Finn 1982). Social judgment-involvement theory takes into consideration a person's level of ego involvement with the attitude concept. It has been established that when a person has low involvement with an issue, his/her latitude of noncommitment (positions along an attitude continuum a person prefers to remain noncommittal) is quite large (Sherif and Sherif 1967). Because of this, it can be difficult to identify accurately that point along an attitude continuum that most closely matches that person's "true" attitude. Though the relationship between ego involvement and certainty has not been empirically tested, one might reasonably expect that many low involvement respondents would also be less certain of their judgments.

Kassarjian (1978), as well as other consumer psychologists have pointed out that many consumer products are of very little personal importance to consumers. Consequently, their level of cognitive involvement with these products and communications about these same products is quite low. Although empirical evidence does not yet exist, it seems reasonable to expect that many people (though surely, not all) are not certain of their attitudes toward low involvement products (perhaps simply because they are of such little concern to the individual that well-developed attitudes are not formed). If this assumption is true, problems should be expected when measuring attitudes toward low involvement products in the traditional manner.

Attitude intensity or a person's strength of conviction in his/her attitude is an issue in attitude measurement that has not received much attention for many years (though the concept is closely related to ego involvement). It has been hypothesized and shown that the correlation between attitude (content) and intensity will yield a J- or U-shaped curve (Suchman 1950; Katz 1944). That is, those with more extreme attitudes (either positive or negative) will be more intense in their attitude position, while those with more neutral or indifferent attitudes will show the least intensity of feeling. Response certainty and intensity are closely related. In several cases, intensity has been operationalized by using a measure of certainty (Suchman 1950). Further discussion of the uses of intensity in attitude research and its relation to response certainty is included in a later section of this paper.

The concept of confidence (or certainty) in attitudes has been an issue in consumer behavior literature in two separate, but related, ways. [Confidence, as used in this study, is treated as being a different and separate construct than self-confidence. Specific self-confidence is a broader concept that is more directly related to a person's self concept while confidence in attitude measurement is a simpler concept referring only to a person's degree of certainty or confidence in their evaluation of beliefs.] Howard and Sheth (1969) propose a consumer's overall confidence in the attitude object (brand) is directly related to purchase intention. Further, Howard and Sheth (1969) as well as others (Cox 1967 and Bauer 1960) have suggested that purchase behavior is affected by a person's degree of certainty about his/her ability to judge the outcome of purchasing a product. Support for both of these uses of confidence was found by Bennett and Harrell (1975).

In combination, the above approaches relating certainty to attitude measurement provide considerable support for continued research and interest in the role of certainty in attitude measurement. While each of these theoretical discussions vary to some degree in terms of why a measure of certainty should be employed in attitude measurement, each clearly suggests that certainty may play an important role in improving predictions of behavior based upon attitudes.

METHODOLOGY

Questionnaires were mailed to a national sample of 1000 households who were members of Market Facts, Inc. Consumer Mail Panel. The sample of 500 males and 500 females was balanced according to latest available census data on four variables: age, annual household income, population density and geographic region. Questionnaires suitable for analysis of the certainty measures were returned by 656 respondents.

The concept under investigation was socially-environmentally responsible behavior. Respondent attitudes were measured by a 40-item Likert scale developed specifically for measuring attitudes toward this concept. Using a procedure similar to that recommended by Churchill (1979), the reliability and validity of the scale were established through the use of: (1) extensive pretesting on relatively large samples, (2) item analysis and factor analysis, (3) the "known groups" technique and a reduced version of Campbell and Fiske's (1970) multitrait-multimethod procedure for construct validation. Scored in the traditional manner (values ranging from 1 to 5), the scale had a coefficient alpha index of reliability of .93. Support for the continued use of this scale has been provided by Henion (1981). For a detailed description of the development of this scale and for further evidence of its reliability and validity, see Antil and Bennett (1979).

The measure of response certainty was obtained through the use of a five-point scale following each item in the attitude scale. Subjects were asked to answer the question, "How certain are you about your decision?" after completing each attitude item. The poles of the scale were labeled "very certain" and "very uncertain." Empirical support for the use of this technique was found by Katz (1944).

The behavioral measure was a self-reported measure consisting of a list of 34 behaviors specifically selected because of their direct relationship to socially-environmentally responsible consumption. Subjects were asked to respond to each behavior by indicating whether the statement was true, false or does not apply. (For example, one behavioral statement was "I belong to a car pool"). In addition to the 34 behaviors of concern, 33 "dummy" behaviors were intermixed in the list to help reduce the likelihood of receiving socially desirable responses. The behavioral measure was placed at the beginning of the questionnaire (first section) thereby reducing the likelihood the remaining questions would influence subjects' responses. A subject's score on the behavioral index was determined by summing those responses which were environmentally compatible behaviors. For a detailed description of the development of the behavior index, see Antil (1978).

RESULTS

Since the research concerning certainty in attitude measurement has been primarily on a theoretical level, little emphasis has been placed on how certainty combines with attitude to form a functional relationship. Thus, this research explored several procedures to incorporate certainty into attitude measurement.

First, an attempt was made to use certainty as an integral variable within the attitude function. To accomplish this, the following multiplicative relationship was employed.

EQUATION  (1)

where:

"C = attitude toward the concept

Bi = the ith belief statement within the 50-item attitude scale

Ci = the ith certainty value for the ith belief

To evaluate the addition of certainty to this model, a comparison of the strength of association between attitude and behavior was made for this model as compared to that obtained with certainty excluded

EQUATION    (2)

Pearson product moment correlations were .55 (p < .001) with the certainty variable, and .56 (p < .001) without certainty. Even without a test of significance, it is obvious that only a minor difference was obtained. Thus, it would seem that the inclusion of certainty in the computation of attitude is not warranted (at least not when used in a multiplicative manner). In a different context, results similar to these were found by Bennett and Harrell (1975).

The usefulness of certainty in attitude research is best illustrated when it is used as a moderator variable [a variable used to divide the sample into more homogeneous groups (Ghiselli 1963)] (Day 1970; Warland and Sample 1973; Bennett and Harrell 1975). To accomplish this, it is necessary to divide the sample into two or more groups based on their degree of certainty. To illustrate the attitude-behavior relation over a range of certainty scores, respondents were divided into five groups of equal size according to their mean certainty scores. The mean correlation between attitude (sum of Likert scale items) and behavior was then plotted for each group.

As is clearly evident from Figure 1, the attitude-behavior correlation increases as respondents become more certain of their responses. For example, the lowest 30 percent of respondents (based on certainty scores) had an average correlation of .403 while the highest 30 percent had an average correlation of .637. Intuitively, one would expect that measuring how certain a person is in his/her attitude would lead to a better measure of that attitude. However, the real significance of this is that the improved measure leads to significantly better correlations between attitude and behavior. Given these results, it is surprising that more attitude research has not included response certainty.

In addition to using the simple mean of respondent confidence scores, another-procedure was attempted. Summation of scores assumes there is a linear relationship between the individual certainty scores, that is, the difference between certainty scores of 1 and 2 is weighted the same as the difference between 4 and 5. It was believed that better results would be obtained by differentially weighting higher certainty scores. This would have the effect of attributing even more influence to those items which the respondent was most certain and proportionately less weight to the items of lower certainty. Through empirical observation, the nonlinear function EQUATION was selected (where x was equal to the respondent's certainty score (1-5)).

FIGURE 1

ATTITUDE-BEHAVIOR CORRELATION BY CERTAINTY GROUPS

Figure 2 presents the results from using the nonlinear function to weight certainty scores. In terms of improving the correlation between attitude and behavior, no significant differences were found. The attitude-behavior correlation for the high median group was .62 while for the low median group it was .45. (Comparable figures when using the linear function were .62 and .48.)

FIGURE 2

ATTITUDE-BEHAVIOR CORRELATIONS BY CERTAINTY GROUPS  (NONLINEAR FUNCTION)

There was, however, a difference in the distribution of correlations across the certainty groupings, the most significant of which was the rather clear break occurring at the median of certainty scores. Such a "natural break" is very useful in dividing the sample into "high" and "low" certainty groups and these groups will be used for the remainder of the analysis.

Although the use of nonlinear weighting of certainty scores did not appreciably improve attitude-behavior correlations in this study, future experimentation is encouraged. Its use is intuitively appealing and applications with other data bases (and different functions) may be fruitful.

Beyond the improvement in the attitude-behavior correlation found when using certainty as a moderator variable, it is useful to examine the distribution of certainty over regions of the attitude scale. Previous research has found a J- or U-shaped relationship between attitude and measures of certainty (Suchman 1950 Gutman 1950; Brim 1955; McCroskey et al. 1967). Such a relationship should exist since a person's intensity of feeling should be strongest at the extreme points of the attitude continuum and decrease as one moves toward the midpoint (Suchman 1950). A similar argument can be made for certainty ratings, however it is also feasible that many (some) of those with more neutral attitudes are also quite certain of their expressed attitude. This position also seems to be related to that proposed by social judgment theory. Sherif and Sherif (1967) report there is considerable evidence that persons with extreme positions on an issue are more likely to be highly involved with the issue. Further, they state the "latitude of rejection increases in size as a function of the extremity of the person's position, and the function is curvilinear for bipolar issues" (p. 118). Since involvement with an issue and how certain a person is in his/her opinions are related, social judgment theory would predict a similar relationship.

Figure 3 and Table 1 present the results obtained from the present research. Figure 3 shows that the results for high certainty respondents do indeed yield a U-shaped relationship while the relationship for the low certainty respondents is just the opposite. (Note though that the mean certainty scores listed in Table 1 do form a U-shaped curve.) Further, from Table 1 it can be seen that 88.6 percent of those with extreme positions (very negative and very positive) were high certainty respondents and of those in the neutral region, 87 percent were low certainty respondents.

FIGURE 3

DISTRIBUTION OF HIGH AND LOW CERTAINTY RESPONDENTS OVER ATTITUDE

[Attitude scale region selection was based upon the results obtained from an intensity function. This function is explained later in this paper.]

TABLE 1

DISTRIBUTION OF HIGH AND LOW CERTAINTY RESPONDENTS OVER ATTITUDE

This difference between high ;and low certainty respondents helps to explain the relationship between certainty and the extent of correlation between attitudes and behavior. One can expect that those with low certainty scores are much more likely to be low involvement respondents (Sherif and Sherif 1967) and have poorly developed attitudes or "non-attitudes" (Converse 1970). Further, as social judgment theory clearly states, those indicating extreme attitude positions are more likely to be much more committed to their attitudes. Thus, one should expect those at the extremes of the attitude continuum to exhibit much higher attitude-behavior consistency while those located near the middle of the attitude continuum to show lower consistency. Although the nature of the data in the present study is not conducive to testing this proposition, Warland and Sample (1973) (whose distribution of certainty ratings were quite similar to those found in the present study) found in their study of voting behavior:

The largest voting difference between the 2 groupings occurred in the positive extreme region of the scale. There was a difference of approximately 58 percent in the incidence of voting behavior between high and low certainty respondents. Those students giving extreme Likert responses but who do not fit the U-shaped curve (the low certainty respondents) did not behave according to their expressed attitudes. (p. 180)

In attitude measurement it often is useful (necessary) to be able to make a summary statement about the overall sentiment of the population measured. Being able to conclude with a high degree of confidence that "x" percent have favorable attitudes and "y" percent have unfavorable attitudes may not always be that easy to do. Can one always automatically assume that the theoretical midpoint of a Likert scale is the dividing point between those with favorable and unfavorable attitudes? If a very large percent of respondents (e.g. 95 percent) have attitude scores in the "positive" ("negative") range of scores, can one conclude that such a large percent of the sample actually have a positive (negative) attitude toward the concept? In order to make such conclusions, the researcher should be confident that the obtained result is an accurate indication of respondents' true attitudes and not due to problems with his/her attitude scale (e.g. negatively or positively biased items, the two ends of the scale are truly opposite in meaning, and the scale consists of items that are a representative sample of all possible items within the attitude domain). In addressing these issues, Suchman (1950) comments:

From these variations in polling results depending upon question wording, and from lack of "expert" agreement as to what is "biased" or "unbiased," it appears quite clear that what is greatly needed is some objective method of dividing the respondents into the same proportions pro and con regardless of question wording. With such a method different opinion pollsters would come out with the same results, even though they should a different questions concerning the same topic. These results would be independent of many of the various "biases" to which public opinion polling today is subject.

Is such a method possible? Can there be an objective method for determining an invariant dichotomization of the population with regard to an opinion or attitude area? (p. 115)

The method proposed by Suchman (1950) involves the use of an intensity function which is derived from measures of certainty or measures of intensity of feeling. He explains

Intensity of feeling is conceived to be strongest at both ends or the scale continuum and to decrease as one moves toward the middle. The scale position with lowest intensity serves to mark off a point of "neutrality" which provides a division of the population into two groups, one positive and the other negative. Furthermore, as a consequence of the scalability or the area, this division does not depend upon the particular sample of questions used. The cutting point is determined objectively and is invariant with respect to wording of questions. (p. 915)

According to Suchman, an intensity function is obtained through the correlation of an attitude scale (which ranks people from high to low on a single continuum) with an intensity scale (which ranks people from strong to weak on a single continuum, e.g. certainty). This should produce a J- or U-shaped function with the turning point (i.e. bottom of J- or U-shaped function) equal to a point of indifference or neutrality (see Katz 1944 for further justification of the existence of a U-shaped intensity curve). As Suchman (1950) illustrates, a most interesting aspect of the intensity function is that for any attitude universe its shape and location of the lowest point is independent of the sample of questions asked, thus, "any sample of questions from that attitude universe will produce the same results" (p. 216).

A slightly modified procedure was used in this research and is as follows:

1. The attitude scale must be divided into several intervals which encompass the entire range of the attitude scores. Each interval should have a reasonable number of subjects (e.g. 30) and there should be enough intervals so that it is possible to obtain a U-shaped function (e.g. 6 or more). [The procedure outlined here differs from that recommended by Suchman (1950) in only the first step. Suchman's attitude scales were scored using a rank order procedure that resulted in "natural" intervals and as such, did not necessitate dividing the attitude continuum into intervals. The procedure and guidelines in Step 1 are recommended by the author.]

2. The median certainty value for each attitude interval is obtained (i.e. for each attitude interval, locate the individual who occupies the median position on certainty and select that certainty score).

3. Determine the certainty percentile for each median certainty score over the entire population (i.e. for the median certainty person in each attitude interval, calculate his/her certainty percentile based upon all certainty scores).

4. The attitude value to be plotted against the median certainty percentile is determined by calculating the percentile of the person(s) at the midpoint of attitude scores within each attitude interval. The percentile is again calculated using scores of all respondents (i.e. for each attitude interval, select the person(s) with the midpoint attitude score within that interval and calculate the percentile for this person based upon all attitude scores.

This procedure results in two percentile scores, one for attitude and the other for certainty.

Percentiles (as opposed to simple median values) are used in order to estimate:

...what the rank on the whole attitude universe would be if indefinitely many questions had been asked.... The percentile metric is used for both content [attitude] and intensity [certainty] so that the people are considered to be arranged from O to 100 percent according to their rank on intensity. (Suchman, p. 220)

Figure 4 illustrates the results obtained from the present research. The resulting U-shaped curve has a distinct minimum point (zero or neutral point) and indicates that one could conclude that just under (allowing for those at neutral point) 81 percent of respondents has favorable attitudes while 19 percent expressed unfavorable attitudes. Although Figure 4 has an obvious zero point. this will not always be the case. In such situations the intensity function is quite useful in estimating not only those expressing positive and negative attitudes, but also those that have fairly neutral attitudes. For example, if the intensity function has a fairly flat bottom all those respondents located in the flat portion of the curve would be considered neutral in opinion. In some situations this can account for a substantial proportion of all respondents (see Suchman 1950 for examples of this). Interpretations such as this can be quite difficult in the absence of an intensity function.

In addition, the development of an intensity function can provide evidence of the overall quality of the attitude measurement device. Specifically, if the intensity function does not yield a J- or U-shaped function it is a strong indication that there may be problems with the attitude scale. In such a situation, consideration should be given to the number of items in the scale (e.g. too few items) and the possibility that the individual items are not each sampling from the same attitude domain (e.g. items are tapping into unrelated constructs).

FIGURE 4

INTENSITY FUNCTION

SUMMARY AND CONCLUSION

This paper has presented several advantages that can accrue through the incorporation of a measure of certainty in attitude measurement. It was clearly shown that when certainty is used as a moderator variable, it is possible to obtain improved correlations between attitude and behavior. While using certainty measures will not increase the predictive ability of attitude scales, its use will allow the researcher to segment the sample of respondents into groups which radically differ in terms of consistency between attitudes and behavior.

This research was not designed to investigate why high certainty respondents showed greater attitude-behavior consistency than low certainty respondents. Speculation was made, however, that response certainty may be related to attitude centrality (Converse 1970), level of ego involvement (Sherif and Sherif 1967) or involvement as used by consumer psychologists (e.g. Houston and Rothschild 1978), the state of development of an individual's attitude (Converse 1970) and attitude intensity (Suchman 1950). While these concepts appear to be highly related, future research which focuses on a better understanding o f their role in attitude measurement is likely to improve our ability to more accurately measure attitudes and gain a better understanding of the attitude-behavior relationship.

The development of an intensity function was shown to be useful in making summary conclusions about the results from an attitude scale as well as illustrate the relation of the level of certainty to the attitude continuum. The U-shaped intensity curve derived from the present research showed that, indeed, respondents at the extremes of the attitude continuum are much more certain of their attitudes than those expressing more neutral attitudes (though most respondents tended to be fairly certain and either positive or negative).

As is evident from the very small amount of research that has used measures of certainty in attitude measurement, the level of interest has been (is) very low. The major question then is, is it advisable to include this variable in attitude measurement? On the negative side of this issue, the major problem would seem to be the apparent necessity to include an additional question for each attitude item. This adds to the length of the questionnaire, can create formatting problems, increases the time to complete a questionnaire, and can add to respondent confusion. The advantages should be clear, with the ability to interpret the results more accurately being the primary advantage. The decision whether to use certainty measures or not would seem to be dictated by: (1) the importance of the research (e.g. how critical is the research in terms of decisions or conclusions to be made), (2) in general, how familiar or knowledgeable are respondents with the attitude construct, and (3) the degree of personal interest or involvement respondents have toward the attitude object. If one can expect a wide variation in the degree of involvement-knowledge among respondents, incorporating certainty would seem advisable. For example, in consumer behavior, many of the topics may be of great interest to the researcher or manufacturer, but of very little concern to many of the consumer-respondents (Kassarjian 1978). Consumer concern over attitudes towards brands of cereal, toothpaste, cake mixes, etc. may vary widely and thus it may be advantageous to segment consumers according to their expressed certainty. Decisions could then be made taking into consideration not only the positive-negative nature of consumer attitudes but also the number (percent) who really seem to care and how these consumers differ from those who do not. In the absence of segmenting by certainty, all consumers would be treated similarly.

One possibility that could make the incorporation of certainty much simpler is the use of a single question (or small number of questions) to measure certainty. Simply asking each respondent a question such as how certain he/she is about the answers (opinions) just given may be adequate and thus eliminate the major problems associated with measuring certainty after each item. (This has been done with some success in attribution theory research: cf. Calder and Burnkrant 1977). This procedure has potential and requires research support before its use can be encouraged.

REFERENCES

Antil, John H. (1978), The Construction and Validation or an Instrument to Measure Socially Responsible Consumption Behavior: A Study of the Socially Responsible Consumer, unpublished doctoral dissertation, The Pennsylvania State University.

Antil, John H., and Bennett, Peter D. (1979), "Construction and Validation of a Scale to Measure Socially Responsible Consumption Behavior," in The Conserver Society, eds. Karl E. Henion and Thomas C. Kinnear, Chicago: American Marketing Association. 51-68.

Bauer, Raymond A. (1960), "Consumer Behavior as Risk Taking," in Dynamic Marketing for a Changing World, R. S. Hancock, ed., Chicago: American Marketing Association.

Bennett, Peter D. and Harrell, Gilbert D. (1975), "The Role of Confidence in Understanding and Predicting Buyers' Attitudes and Purchase Intentions," Journal of Consumer Research, 2, 110-117.

Bettman, James R. (1979), An Information Processing Theory of Consumer Choice, Reading: Addison-Wesley Publishing Company.

Brim, Orville G. (1955), "Attitude Content-Intensity and Probability Expectations," American Sociological Review, 90. 68-78.

Campbell, D. T. and Fiske, D. (1970), "Convergent and Discriminant Validation by the Multitrait-Multimethod Matrix," in Attitude Measurement, ed. G. Summers, Chicago: Rand McNally, 100-22.

Churchill, G. A. (1979), "A Paradigm for Developing Better Measures of Marketing Constructs," Journal of Marketing Research, 16, 64-73.

Converse, Phillip E. (1970), "Attitude and Non-Attitudes: Continuation of a Dialogue," in The Quantitative Analysis of Social Problems, E. R. Tufte, ed., Reading: Addison-Wesley Publishing Company.

Cox, Donald F. (1967), Risk Taking and Information Handling in Consumer Behavior, Boston: Harvard University Press.

Day, George S. (1970), Buyer Attitudes and Brand Choice Behavior, New York: The Free Press.

Finn, David A. (1982), "It is Time to Lay the Low-Involvement Hierarchy to Rest," in An Assessment of Marketing Thought and Practice, eds. Bruce J. Walker et al., Chicago: American Marketing Association, 99-103.

Fishbein, Martin and Ajzen, Icek (1975), Belief, Attitude, Intention and Behavior: An Introduction to Theory and Research, Reading, Mass.: Addison-Wesley.

Ghiselli , Edwin E. (1963), "Moderating Effects and Differential Reliability Validity," Journal of AppLied Psychology, 47, 81-86.

Gutman, Louis (1950), "The Principal Components of Scale Analysis," in .Measurement and Prediction, Samuel Stoffer et al., eds., Princeton: Princeton University Press.

Henion, Karl E. (1981), "Energy Usage and the Conserver Society: Review of the 1979 AMA Conference on Ecological Marketing," Journal of Consumer Research, 8, 339-342.

Houston, Michael J. and Rothschild, Michael L. (1978). "Conceptual and Methodological Perspectives on Involvement," in Research Frontiers in Marketing: Dialogues and Directions, ed. S. C. Jain, Chicago: American Marketing Association, 184-187.

Howard, John A. and Sheth, Jagdish N. (1969), The Theory of Buyer Behavior, New York: John Wiley and Sons.

Kassarjian, Harold H. (1978), "Presidential Address, 1977: Anthropomorphism and Parsimony," in Advances in Consumer Research, vol. 5, Keith Hunt, ed., Ann Arbor: Association for Consumer Research.

Katz, Daniel (1944), "The Measurement of Intensity," in Gauging Public Opinion, Hadley Cantril, ed., Princeton: Princeton University Press.

Krugman, Herbert E. (1965), "The Impact of Television Advertising: Learning Without Involvement," Public Opinion Quarterly, 29, 349-356.

McCroskey, James C., Pichard, Samuel V., and Arnold, William E. (1967), "Attitude Intensity and the Neutral Point on Semantic Differential Scales," Public Opinion Quarterly, 30, 642-645.

Ray, Michael L. (1973), "Marketing Communication and The Hierarchy-of-Effects," in New Models for Mass Communication, ed. Peter Clark, Beverly Hills: Sage Publications. 147-176.

Ryan, C. and Bonfield, E. (1975), "The Fishbein Extended Model and Consumer Behavior." Journal of Consumer Research, 2, 118-136.

Sherif, Carolyn W. and Sherif, Muzafer (1967), "Attitude as the Individual's Own Categories: The Social Judgment-Involvement Approach to Attitude and Attitude Change," in Attitude, Ego,Involvement, and Change, Carolyn W. Sherif and Muzafer Sherif, eds., Westport: Greenwood Press

Sherif, C. W., Sherif, M., and Nebergall, R. E. (1965), Attitude and Attitude Change, Philadelphia: W. B. Saunders.

Suchman, Edward A. (1950), "The Intensity Component in Attitude and Opinion Research," in Measurement and Prediction, Samuel Stoffer et al., eds., Princeton: Princeton University Press.

Warland, Rex H. and Sample, John (1973), "Response Certainty as a Moderator Variable in Attitude Measurement," Rural Sociology, 38, 174-186.

Wicker, A. W. (1969), "Attitudes Versus Actions: The Relationship of Verbal and Overt Behavioral Response to Attitude Objects," Journal of Social Issues, 25, 41-78.

Wilkie, William L. and Pessemier, Edgar A. (1973), "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, 10, 428-41.

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