Opinion Leadership and Innovativeness: Overlap and Validity



Citation:

Jacob Jacoby (1972) ,"Opinion Leadership and Innovativeness: Overlap and Validity", in SV - Proceedings of the Third Annual Conference of the Association for Consumer Research, eds. M. Venkatesan, Chicago, IL : Association for Consumer Research, Pages: 632-649.

Proceedings of the Third Annual Conference of the Association for Consumer Research, 1972      Pages 632-649

OPINION LEADERSHIP AND INNOVATIVENESS: OVERLAP AND VALIDITY

Jacob Jacoby, Purdue University

The construct of opinion leadership has played an important cross-disciplinary role in studies dealing with the adoption and diffusion of products, services, and ideas. Originally described by Lazarsfeld, Berelson, and Gaudet (1948) in their study of the 1940 presidential election, the construct has since been employed by rural sociologists, agricultural economists, communication theorists, social psychologists, and marketers, among others, to study a variety of interpersonal communication phenomena.

While there are various specific measure of opinion leadership, they can all generally trace their origin to one of three different techniques. The self-designating technique is one where the individual is asked to indicate how much of an opinion leader he perceives himself to be. The sociometric approach is one in which all members of a given group are asked to identify those group members considered most influential with respect to the object or idea under consideration. The key informant technique involves first identifying a limited number of people within the group assumed to be knowledgeable regarding the patterns of influence within that group, and then asking them to identify the influentials in that group.

Given the importance and high frequency with which the term opinion leadership appears in the consumer behavior literature, it is surprising to see so few data presented which bear on the interrelationship of these various measurement techniques and on the construct validity of opinion leadership itself. Indeed, the literature seems to contain only two studies which cite original data bearing on the interrelationship of two or all three of these techniques. Katz and Lazarsfeld (1955, pp. 137-161) provide percentage-based evidence to indicate that a high degree of agreement exists between self-designating and sociometric approaches. Rogers and Catarno (1962, p. 441) cite data from an unpublished doctoral dissertation in which all three methods of measuring opinion leadership were applied to 28 dairy farmers. "Self-designating opinion leadership scores were correlated .300 with the number of sociometric choices, and .640 with composite opinion leadership rating by four key informants. The number of sociometric choices was correlated .876 with the composite key informants ratings." However, while data based on assessing opinion leadership with respect to one object or idea do provide an indication of convergent validity, they provide no indication of discriminant validity, and it is the latter which is essential for establishing construct validity (cf. Campbell and Fiske, 1959).

Accordingly, it was the primary purpose of this investigation to apply all three techniques for assessing opinion leadership to several groups of subjects, and to measure such opinion leadership across three different areas of possible influence, so as to assess the convergent and discriminant validity of this construct. This is essentially the multimethod-multitrait approach proposed by Campbell and Fiske (1959). In this regard, construct validity "is primarily concerned with the adequacy of tests as measures of a construct rather than with the adequacy of a construct as determined by the confirmation of theoretically predicted associations with measures of other constructs. We believe that before one can test the relationships between a specific trait and other traits, one must have confidence in one's measure of that trait (1959, p. 100)."

Other purposes of this investigation were: (a) to examine the question of opinion leadership overlap across product categories (cf. King and Summers, 1970); (b) to provide another test of the relationship between opinion leadership and innovativeness (cf. Robertson, 1970, p. 137); and (c) to examine the between-method consistency in identifying innovativeness using the key informant and sociometric techniques.

METHOD

Subjects

Application of the key informant and sociometric techniques is meaningless if all group members do not know, or know of, one another. Consequently, groups which were relatively small and cohesive had to be utilized College fraternities, sororities, and their pledge-classes satisfy these criteria. Thus, the subjects in this investigation were two fraternity pledge-classes (n1 = 13; n2 = 27), and three entire sororities (n3= 60; n4 = 70; n5 = 70) at Purdue University during the Spring 1971 semester. [The assistance of David Giffin and Sue Speaks in collecting these data is appreciatively acknowledged.] The response rate was 100% for Groups 1 and 2; 67% (40/60) for Group 3; 54% (38/70) for Group 4; and substantially less than 50% for Group 5. Accordingly, the data for Group 5 were discarded.

Traits

To assess discriminant validity using the multimethod-multitrait approach, "more than one trait as well as more than one method must be employed in the validity process" (Campbell and Fiske, 1959, p. 81). Accordingly, the two groups of males were examined for opinion leadership in three different product categories: clothing, alcoholic beverages, and LP records. The trait categories for the two groups of females were: clothing, cosmetics, and room decorations.

Instruments

Two slightly different self-designating scales were used for the fraternities and sororities. The scale for the fraternity pledge classes was a modified version of the Rogers and Catarno (1962, pp. 439-440) self-designating opinion leadership scale. The modification attempted to increase reliability and sensitivity by providing more response alternatives for the six questions than were provided in the original scale. A copy of this scale for clothing is included in Appendix A. In all, each fraternity pledge responded to 18 questions--six for each product (i.e., trait). The self-designating scale for the sorority members consisted of seven-item scales for each product patterned after King and Summers (1970, p. 45), and also modified to increase reliability and sensitivity. An example is provided in Appendix A.

The sociometric technique involved asking each member of the group two questions for each product category, and then determining how each member of the group was ranked on these items by his peers. For example, the fraternity pledges were given the following two questions for clothing: "If you were going to a semi-formal dance, which fellow pledges would you go to to get ideas on what to wear?" and "Your parents just gave you some money to buy some new summer clothes. Which fellow pledges would you want to help you select them?" In responding to these questions, each subject rank ordered the five "best" sources he knew of within his group. An index was derived for each subject within each group based upon the average rank he was assigned by his fellow group members. This index was continuous in that, while subjects who never received any votes were always scored as 1.00 and subjects who were always ranked as the "best" source received a score of 6.00, all other combination of rankings resulted in some score (usually not a whole integer) between 1.00 and 6.00.

The key informants in the two sororities were the two presidents and two social chairmen. The key informants in the fraternity pledge-classes were selected on the basis of their sociometric scores. The format for the key informant instrument was comparable to that of the sociometric instrument. The basic difference between the two was that the questions on the sociometric instrument were phrased in the first person ("Whom would you go to when . . . ?"), while those on the key informant instruments were phrased in the third person ("Whom would the majority of the other members go to when . . . ?"). Again, an index was derived for each person for each product category based on his rank on the two questions relating to that product category.

Finally, in the spirit of exploratory investigation, the sociometric and key informant instruments for Groups 3 and 4 also included items designed to assess innovativeness. While the self-designating approach is coming into frequent use as a measure of innovativeness, the sociometric and key informant techniques do not appear among the approaches used to assess innovativeness (cf. King and Ryan, 1971). Therefore, a single question for each of the three product categories was included as a crude measure of innovativeness on both the sociometric and key informant instruments. For example, all participants respond to the question "If you want information about a new cosmetic product, which three girls in the house would you ask?" by citing their best, second best, and third best sources. An innovativeness index was derived for each subject based upon the average rank she was assigned by her sorority sisters. Similarly, the key informants received a single question for each product category--e.g., "Overall, which three girls in the house would be most generally considered to always have the latest look in make-up?"--and responded by providing whom she thought would be the first, second, and third best sources. Again, an innovativeness index was derived.

RESULTS

The multimethod-multitrait matrices for Groups 1-4 are presented in Tables 1-4, respectively, Campbell and Fiske (1959) suggested that one should examine convergent validity before proceeding to assess discriminant validity. The former is accomplished through examining the correlation coefficients along the validity diagonals. These values "should be significantly different from zero and sufficiently large to encourage further examination of validity (p. 82)."

TABLE 1

THE MULTI-METHOD X MULTL-TRAIT MATRIX FOR GROUP 1 (N = 13)

TABLE 2

THE MULTI-METHOD X MULTI-TRAIT MATRIX FOR GROUP 2 (N = 27)

TABLE 3

THE MULTI-METHOD X MULTI-TRAIT MATRIX FOR GROUP 3 (N = 40)

TABLE 4

THE MULTI-METHOD X MULTI-TRAIT MATRIX FOR GROUP 4 (N = 38)

The nine validity diagonal coefficients of Table 1 range from .166 to .965, with a median of .622. Seven of these values are statistically significant at p <.05 or better. The validity diagonal coefficients of Table 2 range from .051 to .591, with a median value of .358. Six of these are significant at p <.05 or better, and a seventh nearly so (p <.06). Table 3 contains validity diagonal coefficients which range from .208 to .591, with a median value of .414. Eight of these coefficients are statistically significant at p <.05 or better. Finally, the validity diagonal coefficients of Table 4 range from .043 to .889, with a median value of .421. Seven of these coefficients are statistically significant at p <.02 or better.

Thus, approximately 80 x of these coefficients (28 out of 36) reach acceptable levels of statistical significance and a 29th nearly so (p <.06). The median value across all 36 coefficients is .440. These values are surprisingly high in view of the limits that the reliability of an instrument places on its validity. Rogers and Catarno (1962, p. 446) reported a split-half reliability of .703 for their self-designating scale. While the reliabilities of the specific instruments used in the investigation were not established, it is very probable that they were not much higher. It would appear, therefore, that convergent validity has been adequately established.

Three criteria must be satisfied in order to demonstrate discriminant validity. First, "a validity value for a variable should be higher than the correlation obtained between that variable and any other variable having neither trait nor method in common (Campbell & Fiske, 1959, p. 82)." This requires that each validity diagonal value be higher than the value lying in its column and row in the heteromethod-heterotrait triangles. For example, the validity diagonal coefficient based on the Self Designating and Key Informant assessment of clothing opinion leadership in Table 1 is .545. It is higher than all eight of the heteromethod-heterotrait coefficients lying in its column (.247, .449, .248, .177) and row (.335, .038, -.196, .519). Each of the nine validity diagonal coefficients was involved in eight direct comparisons. Across all groups, approximately 72% (207/288) of these comparisons satisfied the first discriminant validity criterion. More specifically, 65/72 did so in Group 1, 26/72 in Group 2, 56/72 in Group 3, and 60/72 in Group 4.

The second criterion for establishing discriminant validity requires "that a variable correlate higher with an independent effort to measure the same trait than with measures designed to get at different traits which happen to employ the same method (Campbell and Fiske, 1959, p. 83)." This requires that each value along the validity diagonals be compared against the two values in each of the two heterotrait-monomethod triangles which use the same two methods, but which depict how well that trait correlates with other traits using the same methods. As an example, consider the first validity diagonal value in Table 1, r - .759. This coefficient represents the correlation between two methods (self-designating and sociometric) of assessing the same trait (opinion leadership with respect to alcoholic beverages). It must, therefore, be demonstrated that this value is higher than correlations between that trait and the other two traits using either of the two methods in question. This criterion is completely satisfied in this instance inasmuch as the correlations between being an alcoholic beverage opinion leader and either a record or clothing opinion leader are only .065 and -.021, respectively, for the self-designating monomethod-heterotrait block, and only .283 and .153, respectively, in the sociometric monomethod-heterotrait block.

There are 36 comparisons (9 validity diagonal coefficients X 2 values in each of the appropriate monomethod-heterotrait blocks) to be made for each of Tables 1-4. Thirty such comparisons in Table 1 satisfied this second criterion, 15 in Table 2, 13 in Table 3, and 22 in Table 4. Overall, 80 of 144 comparisons (55.44%) satisfied this second criterion. These results are considerably better than most such results reported in the behavioral science literature.

The third and final criterion necessary for providing evidence of discriminant validity "is that the same pattern of trait interrelationships be shown in all of the heterotrait triangles of both the monomethod and heteromethod blocks (Campbell and Fiske, 1959, p 83)". Accordingly, the three coefficients in each of the nine triangles were rank ordered for each Group, and Kendall's coefficient of concordance (W) was computed to determine the extent of agreement among the nine sets of ranks within each Group (Siegel, 1956, pp. 229-238). Tables 5-8 present these data. In summary, the values of S for Group 1 (78) ant Group 4 (131) were significant at p <.01, but failed to reach traditional levels of statistical significance in Groups 2 and 3.

TABLE 5

KENDALL'S COEFFICIENT OF CONCORDANCE (W) FOR GROUP 1

TABLE 6

KENDALL'S COEFFICIENT OF CONCORDANCE (W) FOR GROUP 2

TABLE 7

KENDALL'S COEFFICIENT OF CONCORDANCE (W) FOR GROUP 3

TABLE 8

KENDALL'S COEFFICIENT OF CONCORDANCE (W) FOR GROUP 4

Table 9 summarizes the results of the four validity tests for each of the four groups. It appears as if construct validity was adequately established in Groups 1 and 4 but not in Groups 2 and 3.

TABLE 9

SUMMARY OF TESTS FOR ESTABLISHING CONSTRUCT VALIDITY OF OPINION LEADERSHIP

Examination of the monomethod-heterotrait triangles of Tables l-4 reveals that, overall, these coefficients tend to be positive and significant. With respect to Group 1, seven of nine values are positive and one of these is significant. All nine coefficients of Table 2 are positive and five of these are significant. Similarly, all nine coefficients of Table 3 are positive and seven of these are statistically significant. Finally, eight of the nine coefficients in Table 4 are positive of which four are significant. Overall, 33 of 36 coefficients were positive and nearly half (17/36) were significant.

There are several implications which stem from these immediately preceding data. One such implication is that they reflect strong methods variance. A second implication is that these data offer moderate to strong support for the contention that opinion leadership overlap exists across product categories (cf. King and Summers, 1970; Marcus and Bauer, 1964). Moreover, the more similar the product categories, the greater the degree of overlap. As examples, there is greater overlap within the sororities between clothing and cosmetic opinion leaders than between clothing and room decoration opinion leaders, or between cosmetic and room decoration opinion leaders (cf. Tables 7 and 8). This is the same pattern of results obtained in Myers and Robertson's (1972) study of 246 Los Angeles housewives. While none of the three products used with the fraternity pledges are similar to each other, it is still obvious that there is less opinion leadership overlap between alcoholic beverage and record opinion leaders than between alcoholic beverage and clothing or clothing and record opinion leaders (cf. Tables 5 and 6).

A third, and probably more significant, implication to be derived from these data is that, to the extent that there is opinion leader overlap, the three "traits" selected to assess the validity of opinion leadership as a construct (i.e., opinion leadership in three specific product categories) are not truly independent and, as such, provide a more stringent test of construct validity. That is, had the sociometric, self-designating, and key informant methods been used to assess opinion leadership (for clothing), brand loyalty (in dentifrices), and deal proneness (in detergent) rather than having assessed opinion leadership for clothing, opinion leadership for cosmetics, and opinion leadership for room decorations, it would probably have been easier to establish validity for the construct of opinion leadership. Under these circumstances, the degree of construct validity actually obtained must be considered very encouraging.

Table 10 contains the intercorrelations among the three methods of assessing overall opinion leadership for all four groups. These coefficients are based upon "general opinion leadership" scores developed from summing each subject's separate opinion leadership scores across all three product categories. The fact that these coefficients are considerably higher than when opinion leadership was considered on a product-by-product basis (median value of .605 versus .440) also tends to support the notion of opinion leadership overlap. The values in Table 10 would have been considerably lower had opinion leadership correlated either negatively or negligibly across product categories.

TABLE 10

INTERCORRELATION OF METHODS USING OVERALL OPINION LEADERSHIP SCORES

Table 11 presents the intercorrelations between the opinion leadership index and the single item innovativeness scores for the three products employed with the sorority women. These data add support to the findings of others (cf. Robertson, 1970, pp. 136-137; Myers and Robertson, 1971; Rogers and Stanfield, 1968; Tigert and Arnold, 1971) that a positive relationship exists between opinion leadership and innovativenss. Indeed, the coefficients seem incredibly high (median = .800). However, the fact that they are based upon only a single question dictates that they be interpreted with extreme caution.

TABLE 11

CORRELATIONS BETWEEN "INNOVATIVENESS" AND "OPINION LEADERSHIP" SCORES

Table 12 presents the multimethod-multitrait matrices for the single item innovation indices continued within the sociometric and key informant instruments administered to Groups 3 and 4* Both matrices provide adequate evidence of convergent validity. The six coefficients comprising the two validity diagonals range from .509 to .851, and all are significant at p <.01. Twenty-three of twenty-four tests satisfy the first criterion of discriminant validity, viz., that the values on the validity diagonal be higher than the coefficients in their respective rows and columns. The only exception occurs in regard to assessing clothing innovativeness in Group 4. The second criterion (as described above) requires that 12 comparisons be made for each of these two matrices. All 12 comparisons of G.roup 3 satisfy this criterion while 11 of the 12 comparisons do so in Group 4. The third discriminant validity criterion requires that the same pattern of relationships be manifested within all triangles. A precisely identical pattern is manifested in all triangles of Group 4, while Group 3 manifests only one minor inversion: the clothing-room decoration and cosmetics-room decoration correlation coefficients exchange second and third place ranks in one of the four triangles. While the tables for critical values of S in the Kendall coefficient of concordance (cf. Siegel, 1956, p. 286) do not contain values for seven or fewer rankings of sets of three coefficients, consideration of the values in Table 12 suggests that both Ws would be significant at p <.01 had values for these conditions been provided. In sum, the data in Table 12 suggest a considerable amount of construct validity for the notion of innovativeness as assessed by the crude, single-item measures employed in this investigation.

CONCLUSIONS

Despite the fact that the range of values was probably restricted by the nature of the subjects employed (i.e., all were young, relatively literate, members of the "Greek" community on the same midwestern campus), the data collected in this investigation argue for acceptance of the following conclusions:

1. Simultaneously applying the self-designating, sociometric, and key informant methods of measuring opinion leadership to relatively cohesive groups reveals that there is a substantial degree of construct validity for the notion of opinion leadership.

2. Opinion leadership tends to overlap different product categories, and the more similar the product categories, the greater the degree of overlap. These findings are consistent with the results of other mono-method investigations of opinion leadership (e.g., King ant Summers, 1970; Marcus and Bauer, 1964; Myers and Robertson, 1972; Montgomery and Silk, 1971).

3. Opinion leadership is positively correlated with innovativeness. Again, this is consistent with the results of other investigation (e.g., Myers and Robertson, 1972; Robertson, 1970, pp. 136-137; Rogers and Stanfield, 1968; Tigert and Arnold, 1971).

4. Innovativeness, which can be assessed sociometrically and via key informants, would also appear to possess a high degree of construct validity.

TABLE 12

MULTIMETHOD-MULTITRAIT MATRICES FOR GROUPS 3 AND 4 ON INNOVATIVENESS

REFERENCES

Campbell, D. T., & Fiske, D. W. Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin, 1959, 56, 81-105.

Katz, E., & Lazarsfeld, P. F. Personal influence: The part played by people in the flow of mass communications. Glencoe, Illinois: Free Press, 1955.

King, C. W., & Ryan, G. E. Identifying the innovator as a consumer change agent. In D. Gardner (Ed.), Proceedings, 2nd Annual Conference, Association for Consumer Research, 1971, 446-451.

King, C. W., & Summers, J. O. Overlap of opinion leadership. across consumer product categories. Journal of Marketing Research, 1970, 7, 43-50.

Lazarsfeld, P. F., Berelson, B., & Gaudet, H. The people's choice. 2nd ed. New York: Columbia University Press, 1948.

Marcus, A. S., & Bauer, R. A. Yes: There are generalized opinion leaders. Public Opinion Quarterly, 1964, 235 628-632.

Myers, J. H., & Robertson, T. S. Dimensions of opinion leadership. Journal of Marketing Research, 1972, 9, 41-46.

Montgomery, D. B., & Silk, A. J. Clusters of consumer interests and opinion leader spheres of influence. Journal of Marketing Research, 1971, 8, 317-321.

Robertson, T. S. Consumer behavior. Glenview, Illinois: Scott Foresman and Company, 1970.

Rogers, E. M., & Catarno, D. G. Methods of measuring opinion leadership. Public Opinion QuarterlY, 1962, 26, 435-441.

Rogers, E. M., & Stanfield, J. E. Adoption and diffusion of new products: Emerging generalizations and hypotheses. In P. M. Bass, C. W. King, & E. A. Pessemier (Eds.), Application of the sciences to marketing management. New York: Wiley, 1968, 227-250.

Siegel, S. Nonparametric statistics for the behavioral sciences. New York: McGraw-Hill, 1956.

Tigert, D. J., & Arnold, S. J. Profiling self-designated opinion leaders and self-designated innovators through life style research. In D. Gardner (Ed.), Proceedings, 2nd Annual Conference, Association for Consumer Research, 1971, 425-445.

APPENDIX A

SELF-DESIGNATING OPINION LEADERSHIP INSTRUMENT FOR GROUPS 1 AND 2

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

Authors

Jacob Jacoby, Purdue University



Volume

SV - Proceedings of the Third Annual Conference of the Association for Consumer Research | 1972



Share Proceeding

Featured papers

See More

Featured

How Passive Form Messages in CSR Advertisement Improve Consumer Reaction to the Campaign

Taehoon Park, University of South Carolina, USA
Anastasiya Pocheptsova Ghosh, University of Arizona, USA
Elise Chandon Ince, University of South Carolina, USA

Read More

Featured

Communicating a Promise of Change: Visual Steps Enhance Process Imagery

Luca Cian, University of Virginia, USA
aradhna krishna, University of Michigan, USA
Chiara Longoni, Boston University, USA

Read More

Featured

Taking a Leaf out of my Review: The Asymmetrical Link between Linguistic Similarity and Attitude Certainty for Writers and Readers of Product Reviews

Ann Kronrod, University of Massachusetts, USA
Yakov Bart, Northeastern University, USA

Read More

Engage with Us

Becoming an Association for Consumer Research member is simple. Membership in ACR is relatively inexpensive, but brings significant benefits to its members.