# Sparks and Tucker Revisited: a Reanalysis and Replication

^{[ to cite ]:}

Charles M. Schaninger and V. Parker Lessig (1978) ,"Sparks and Tucker Revisited: a Reanalysis and Replication", in NA - Advances in Consumer Research Volume 05, eds. Kent Hunt, Ann Abor, MI : Association for Consumer Research, Pages: 295-299.

^{[ direct url ]:}

http://acrwebsite.org/volumes/9438/volumes/v05/NA-05

The personality and product usage measures employed in the Sparks and Tucker (1971) study were replicated on a sample of 142 male college students at a large Midwestern university. A comparison of the simple correlation matrices between the two sets of data revealed only limited similarity. Both sets of data revealed canonical roots of similar magnitude and significance, although only limited similarity of canonical loadings among the canonical functions was demonstrated.

INTRODUCTION

Within the field of personality and product usage, most of the research appearing since the late 1960's has been strongly influenced by Sparks and Tucker's (1971) study. Prior to this and Kernan's (1968) study, most of the work in this area had focused on either simple univariate analyses, or on single criterion, multiple predictor studies. Canonical correlation analyses, cluster analyses, and multiple discriminant analyses have been used in a large number of studies (Alpert, 1972; Greeno, Sommers, and Kernan, 1973; Darden and Reynolds, 1974; Villani, 1975). Essentially, these studies suggest the existence of "molar personality types" -- i.e., patterns of product usage across product classes associated with corresponding personality profiles. Personality is viewed in a gestalt sense, with various combined multivariate approaches used to identify the different molar types. Many studies have utilized similar multivariate-statistical approaches with AIO/lifestyle/psychographic data, to identify "molar types" or segments in the market. Books by King and Tigert (1971), and Wells (1974), as well as papers by Wells (1975) and Kassarjian and Sheffett (1975), describe this vein of research.

The Sparks and Tucker paper had a strong impact because it took a broadened conceptual approach toward the area of personality and product usage, and because it's conclusions suggested the existence of much stronger and more meaningful relationships than had been found in previous investigations.

A number of authors (Kollatt, Blackwell and Engel, 1972; Engel, Kollatt and Blackwell, 1973; and Robertson and Ward, 1973) have stressed the need for replication of consumer behavior studies, particularly for studies which have a significant impact and which used limited samples. Our objective was to replicate (using new data) and to reexamine (using Sparks and Tucker's original data) the Sparks and Tucker study. We were specifically interested in examining the degree to which the original and the replicated studies yield comparable results.

METHODS

The subjects for the replication were 155 male undergraduate college students enrolled in introductory marketing courses at a large Midwestern university. Data were collected for the same eight personality variables and seventeen product-usage variables used by Sparks and Tucker. Of the 155 subjects comprising our total sample, 142 accurately completed all questions. This sample is roughly comparable to that of Sparks and Tucker's original study -- although the samples do come from universities located in different geographical areas and from different time periods.

We obtained and reanalyzed the raw data for the 173 subjects completing all questions for the original Sparks and Tucker study. [These data were made available through the courtesy of Professor Sparks.] We found a number of minor discrepancies between the correlations based upon reanalyzing their data and those reported in the original article. Evidently their calculations were based on pairwise deletions of missing data on their entire sample of 190 subjects, rather than on the reduced sample of fully completed forms for the 173 subjects serving as our reanalysis sample.

The relationships between personality and product usage were examined using the two approaches presented by Sparks and Tucker: an examination of the simple correlations between each personality trait and product usage score, and an examination of the canonical correlation analysis which focused on the overall association between the two sets of variables (product usage and personality). For both the original and replication samples, the overall strength of association between the two sets of variables was examined by Bartlett's test of canonical correlation significance, by Veldman's test of canonical correlation significance, and by Stuart and Love's (1968) redundancy measure.

Finally, the nature of the relationships found between the personality and product usage measures for the two samples was compared by examining the factor loadings associated with each set of linear canonical functions.

RESULTS

Simple Correlations

From an examination of their correlation matrix, Sparks and Tucker reported "some weak and spotty relationships between personality traits and particular product use." They found 18 of the 136 cross correlations to be significant at the 0.05 level. Our reanalysis of Sparks and Tucker's data (n = 173) also produced 18 significant correlations; 13 of these were the same as Sparks and Tucker found. Only 12 correlations were significant on the replication sample; of these only five were the same as those of the original study. These five were between the following pairs: ascendancy and use of alcoholic beverages, ascendancy and fashion adoption, sociability and use of alcoholic beverages, sociability and fashion adoption, and between cautiousness and use of alcoholic beverages.

Multivariate Relationships

Table 1 presents the first three canonical roots for our reanalysis of Sparks and Tucker's data (n = 173), for the replication study (n = 142), and those reported in Sparks and Tucker's article. We note that in both the original Sparks and Tucker study and the reanalysis of their data by Lambert and Durand (1975), it was concluded that the first three roots are significant. However, the third root in both studies was significant at the 0.01 level and not at the 0.05 level. Based upon their observations, Sparks and Tucker interpreted the first three sets of linear canonical functions. It should be noted that the studies by Sparks and Tucker and by Lambert and Durand applied the significance test developed by Veldman (1967), rather than Bartlett's (1941) test. When Veldman's test was applied to Sparks and Tucker's data (n = 173, not 190) and to our replication data, only the first two roots were significant. Examination of the Bartlett's test revealed only the first canonical root to be significant for each set of data. (These chi-square values and their corresponding degrees of freedom are also reported in Table 1.)

The discrepancy between the Veldman and the Bartlett procedures concerning the number of significant canonical roots should be noted. The correspondence between these two tests has been presented by Alpert, Peterson, and Martin (1975). Essentially, Veldman's test partitions the first Bartlett test chi-square value (of the significance of the first and remaining sets of canonical functions) to calculate the relative chi-square contribution attributable to each of the roots. Disregarding the significance testing procedure employed, both sets of data were comparable with regard to the number of significant canonical roots.

Although both Bartlett's test and Veldman's test provide tests of the statistical significance of the association between two sets of variables, little additional information is provided. Stewart and Love's redundancy test was therefore applied to measure the amount of variation in the set of personality scores. Table 1 also shows the redundancy scores obtained on the Sparks and Tucker data and on the replication data. These scores indicate that 7.37 percent of the variation of product usage is associated with variation in personality for the Sparks and Tucker data, and 8.28 percent for the replication data. The reader is referred to Lambert and Durand (1975) and to Alpert, Peterson, and Martin (1975) for a discussion of this multivariate squared multiple correlation measure. It should be noted that the more complex mathematical form for its calculation simplifies to the mathematical average of the multiple R^{2} for each criterion variable with the set of predictors.

Nature of the Relationships

Table 2 presents the factor loadings for the first and the second set of canonical functions for our reanalysis of Sparks and Tucker's data and for our replication sample. It should be noted that loadings represent correlations between the original variables and the canonical variate scores, and not the canonical weights (coefficients) used to multiply standardized variable scores calculating canonical variate scores. In several instances, both the direction and magnitude of the weights were different from the loadings for the reanalysis and for the replication data. When colinearity existed, the weights tended to be lower than the loadings in magnitude and when a given variable had a suppressor effect, its weight was opposite in direction to the loading.

An examination of the first set of canonical loadings for the two samples demonstrated that the contributions made by the original variables in explaining the first set of canonical relationships differ. An interpretation of these loadings for the Sparks and Tucker data suggests a pattern of heavy usage of cologne and alcohol, and high fashion adoption associated with high sociability and ascendancy, and with low responsibility and cautiousness. The comparable first set of canonical functions for the replication sample suggested a pattern of low shampoo usage and high coffee usage associated with higher scores on original thinking, cautiousness, and personal relations. Both conclusions may be viewed as intuitively appealing, but neither could be described as representing the same basic underlying relationship.

The second set of linear functions from the Sparks and Tucker data demonstrated a product usage pattern consisting of heavy usage of mouthwash, headache remedies, and coffee associated with lower emotional stability, vigor, and original thinking, and higher cautiousness scores. This interpretation, based on loadings, is basically consistent with that given by Sparks and Tucker, based upon their examination of their canonical coefficients (weights). It should be noted that the test of emotional stability, from the Gordon Personal Profile, tends to load highly on items related to nervousness, rather than to those which would indicate emotional instability in a clinical sense.

The pattern of loadings for the second set of canonical functions for the replication data was also dissimilar to the second function for the Sparks and Tucker data, although relatively similar to the first function for their data. A product usage pattern of heavier consumption of alcoholic beverages, chewing gum, and fashion adoption was associated with lower cautiousness and higher sociability and ascendancy. Although there are differences between the second canonical function in the replication data and the first for the Sparks and Tucker data, a similar underlying pattern tends to emerge.

Although the roots were not statistically significant, there was also some degree of similarity between the third root of the replication data and Sparks and Tucker's second root; however, the similarities were not as clear-cut as those between the second and first (respectively) roots. Cross loadings were also calculated (the correlation between the individual predictor variables and the criterion canonical variate scores and vice versa). These were, as expected, generally lower, yet basically comparable to the original loadings for both data sets. All of the variables mentioned demonstrated significant loadings (correlations) at the .05 level, and the great majority of the cross loadings were also significant at this level. Although the significance tests of the loadings and cross loadings cannot be interpreted as significant correlations between independent variables, their relative significance levels merit examination in lieu of any other significance measures.

SUMMARY

A comparison of the findings obtained from the reanalysis of Sparks and Tucker's data and the same analysis on the replication data reveals several similarities worth noting. First, as shown in Table 1, the magnitude of the first three canonical roots associated with the two sets of data is very similar. Second, for both sets of data, according to Bartlett's test only the first set of canonical functions is statistically significant. However, using Veldman's approach, the first two sets of canonical functions are shown to be significant for both sets of data. Finally, strong similarities are noted in the redundancy scores obtained between product usage measures and personality traits. The redundancies measured from the reanalysis of Sparks and Tucker's data and from the replication data were 7.37 and 8.28, respectively. These figures compare favorably to the redundancy of 6.2 found by Lambert and Durand.

A comparison of the matrices of simple correlations between product usage scores and personality measures for the two sets of data revealed a similar number of significant correlations, although less than half of these significant correlations were for the same pairs of variables. Direct similarities were not noted between the two sets of data when the factor loadings for the first two canonical functions were compared, although the second canonical function for the replication data revealed a pattern similar to that demonstrated in the first function of the Sparks and Tucker data.

The differences in the simple correlations between those reported in Sparks and Tucker's article, based on the full sample of 190 and those found by reanalyzing the data for the 173 with complete data, as well as the evidence of instability of canonical weights found by Lambert and Durand, suggest that the differences in the findings are due more to violation of the statistical assumptions underlying canonical analysis and to random error in the relatively weak relationships found for these small sample steps than to differences between the two samples.

Based upon the consistency of results across a number of different studies, it does appear that a statistically significant relationship exists between product usage and personality traits when these measures are treated as a whole, i.e., multivariately. Given the magnitude of the redundancy indices, ranging from 6.2 to 8.28 percent, it cannot be concluded that this relationship is significant in a practical sense. Although the relationships demonstrated by Sparks and Tucker's first canonical function do appear to be supported by the loadings of the second function of the replication data, there were still many inconsistencies between the loadings and percentages across the different canonical functions for the two sets of data. Generalizations as to the specific nature of relationships between personality traits and product usage are questionable, and probably should not be made without stronger evidence than that which has been obtained from an unreplicated study.

REFERENCES

Mark I. Alpert, "Personality and the Determinants of Product Choice," __Journal of Marketing Research__, 9 (February 1972), 89-92.

Mark I. Alpert, Robert A. Peterson and Warren S. Martin, "Testing the Significance of Canonical Correlations,'' __Combined Proceedings__, American Marketing Association, 1975, 117-19.

M. S. Bartlett, "The Statistical Significance of Canonical Correlations," __Biometrika__, 32 (January 1941), 29-38.

William R. Darden and Fred D. Reynolds, "Backward Profiling of Male Innovators," __Journal of Marketing Research__, 11 (February 1974), 79-85.

James F. Engell, David T. Kollatt and Robert D. Blackwell, __Consumer Behavior__ (2nd edition), New York: Holt, Rinehart & Winston, 1973.

Daniel W. Greeno, Montrose S. Sommers and Jerome B. Kernan, "Personality and Implicit Behavior Patterns," __Journal of Marketing Research__, 10 (February 1973), 63-69.

Harold H. Kassarjian, "Personality and Consumer Behavior: A Review," __Journal of Marketing Research__, 8 (November 1971), 409-418.

Harold H. Kassarjian and Mary J. Sheffett, "Personality and Consumer Behavior: One More Time," __Combined Proceedings__, American Marketing Association, 1975, 197-201.

Jerome Kernan, "Choice Criteria, Decision Behavior, and Personality," __Journal of Marketing Research__, 5 (May 1968), 155-64.

Charles W. King and Douglas J. Tigert (eds.), __Attitude Research Reaches New Heights__, Chicago: American Marketing Association, 1971.

David T. Kollatt, Robert D. Blackwell and James F. Engel, "The Current Status of Consumer Behavior Research's Development during the 1968-72 Period," __Association for Consumer Research Proceedings__, 1972, 576-85.

Zarrel V. Lambert and Richard M. Durand, "Some Precautions in Using Canonical Analysis," __Journal of Marketing Research__, 12 (November 1975), 468-75.

Thomas S. Robertson and Scott Ward, "Consumer Behavior Research: Promise and Prospects," in __Consumer Behavior: Theoretical Sources__, T. S. Robertson and S. Ward (eds.), Englewood Cliffs, New Jersey: Prentice-Hall, 1973, 3-42.

David L. Sparks and W. T. Tucker, "A Multivariate Analysis of Personality and Product Use," __Journal of Marketing Research__, 8 (February 1971), 67-70.

Douglas Stewart and William Love, "A General Canonical Correlation Index," __Psychological Bulletin__, 70 (September 1968), 160-3.

Kathryn E. Villani, "Personality/Life Style and Television Viewing Behavior," __Journal of Marketing Research__, 12 (November 1975), 432-39.

William D. Wells, "Psychographics: A Critical Review," __Journal of Marketing Research__, 12 (May 1975), 196-213.

William D. Wells (ed.), __Life Style and Psychographics__, Chicago: American Marketing Association, 1974.

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

Tweet
window.twttr = (function (d, s, id) { var js, fjs = d.getElementsByTagName(s)[0], t = window.twttr || {}; if (d.getElementById(id)) return; js = d.createElement(s); js.id = id; js.src = "https://platform.twitter.com/widgets.js"; fjs.parentNode.insertBefore(js, fjs); t._e = []; t.ready = function (f) { t._e.push(f); }; return t; } (document, "script", "twitter-wjs"));