The Dimensionality of Involvement: an Empirical Test

ABSTRACT - This study was an additional attempt at providing further empirical evidence of the dimensionality of involvement. This research extended previous investigations of this issue by examining involvement dimensionality across multiple products using exploratory and confirmatory factor analyses. Lastovicka and Gardner's (1979) involvement scale served as the involvement measure due to its use in previous research and since it was developed using multiple products. A four dimensional solution provided the best representation of involvement as tapped by this scale. Results suggest the first factor may represent involvement across products while other factors appear to be product specific.

Citation:

Thomas D. Jensen, Les Carlson, and Carolyn Tripp (1989) ,"The Dimensionality of Involvement: an Empirical Test", in NA - Advances in Consumer Research Volume 16, eds. Thomas K. Srull, Provo, UT : Association for Consumer Research, Pages: 680-689.

Advances in Consumer Research Volume 16, 1989      Pages 680-689

THE DIMENSIONALITY OF INVOLVEMENT: AN EMPIRICAL TEST

Thomas D. Jensen, University of Arkansas

Les Carlson, University of Arkansas

Carolyn Tripp, University of Arkansas

ABSTRACT -

This study was an additional attempt at providing further empirical evidence of the dimensionality of involvement. This research extended previous investigations of this issue by examining involvement dimensionality across multiple products using exploratory and confirmatory factor analyses. Lastovicka and Gardner's (1979) involvement scale served as the involvement measure due to its use in previous research and since it was developed using multiple products. A four dimensional solution provided the best representation of involvement as tapped by this scale. Results suggest the first factor may represent involvement across products while other factors appear to be product specific.

During the past decade a number of scales have been developed purporting to measure involvement and the dimensions underlying the construct (i.e., Bloch 1981; Lastovicka & Gardner 1979; Laurent & Kapferer 1985; Slama & Tashchian 1985; Traylor & Joseph 1984; Zaichkowsky 1985). With the exception of Shimp and Sharma's (1983) test of Bloch's (1981) automobile involvement scale, no independent attempts at replication of the involvement scales nor the dimensionality have been undertaken. Furthermore, although Shimp and Sharma extended the automobile scale via the utilization of a fairly large nonstudent sample and refinement of the scale via item and factor (dimension) reductions using confirmatory factor analyses, they investigated only a single product: automobiles. As suggested by Shimp and Sharma, additional studies are clearly warranted that independently ascertain the dimensionality of involvement scales across multiple products.

The present study attempts to partially remedy the above situation by testing the dimensionality of an involvement scale across products using procedures similar to those utilized by Shimp and Sharma (1983). Specifically, the present study addresses the replication and dimensionality issues via the utilization of confirmatory factor analysis procedures for three products using Lastovicka and Gardner's (1979) involvement scale. In addition to providing evidence for the number of dimensions in Lastovicka and Gardner's scale, this study also addresses whether involvement is unidimensional, multidimensional, or multidimensional with only a single consistent dimension across product categories. Finally, the relations between involvement dimensions are examined.

EMPIRICAL DIMENSIONS OF INVOLVEMENT

In explicating the empirical dimensions of involvement it is important to emphasize the fact that this study is not concerned with the conceptual definitions and dimensions, nor types of involvement. Rather, this study focuses on the empirically determined dimensions of existing involvement scales.

Six basic and widely disseminated involvement scales have been developed (i.e., Bloch 1981; Lastovicka & Gardner 1979; Laurent & Kapferer 1985; Slama & Tashchian 1985; Traylor & Joseph 1984; Zaichkowsky 1985). With one exception, all of these scales have utilized Likert-type response formats using from 6 (Traylor & Joseph 1984) to 33 items (Slama & Tashchian 19852. Zaichkowsky (1985), on the other hand, argued that Likert scale items were problematic "because items that seemed to be appropriate for frequently purchased goods did not seem to apply to durable goods and vie versa" (p. 342). Hence, Zaichkowsky developed a 20 item semantic differential scale to measure involvement.

The number of empirically derived involvement dimensions using these scales has ranged from one (e.g., Zaichkowsky 1985; Traylor & Joseph 1984) to six dimensions (Bloch 1981). In the latter case, as noted previously, Shimp and Sharma (1983) found that Bloch's 17 item, six-dimension scale could be reliably reduced to eight items with two dimensions representative of emotional/personal (enduring) and social status (situational) involvement. Bloch had implied a similar configuration in stating that three of his original dimensions tapped centrality while the other three dimensions seemed to tap product interest.

Given the single product, automobiles, tested by Bloch (1981) and, subsequently, Shimp and Sharma (1983), it is important to contrast their findings with those of Zaichkowsky (1985) and Traylor and Joseph (1984) using multiple products. These authors reported that a second reliable dimension could not be ascertained when testing their scales across products. Although these authors allowed for involvement to be multidimensional (see especially Traylor & Joseph 1985), in both scales the lack of a second reliable dimension suggests that (a) involvement may be unidimensional or (b) multidimensional with subsequent dimensions beyond unity being product specific: a notion similar to Bloch's (1981) rationale in originally developing the automobile involvement scale.

Lastovicka and Gardner (1979), on the other hand, reported three dimensions in their 22 item scale: familiarity, commitment, and normative importance. The latter two dimensions appear to be consistent with Shimp and Sharma's (1983) emotional/personal and social status involvement, respectively. Lastovicka and Gardner (1979) suggested that "familiarity is independent of the components of involvement" (p. 68). Unfortunately, Lastovicka and Gardner did not compare the factor loadings of their items across the 14 products they investigated as did Zaichkowsky (1985) and Traylor and Joseph (1984). Hence, the notion of unidimensional, multidimensional, or multidimensional with only a single common dimension across products could not be ascertained.

The remaining two involvement scales represent unique attempts at measuring involvement. Slama and Tashchian's (1985) scale was designed to tap general purchasing involvement as opposed to product involvement. These authors did not attempt to directly ascertain the empirical dimensions in their scale. Rather, they highlighted the attitudinal and behavioral components that might determine involvement. Laurent and Kapferer (1985) attempted -to identify and measure the antecedents of involvement. Their factor analysis suggested four antecedent dimensions: the importance of the product and consequences of making a wrong purchase (imporisk), the probability of making a wrong purchase (risk probability), the symbolic value of the product (sign value), and the emotional value of the product (hedonic value). Although not necessarily directly comparable given their purposes, Laurent and Kapferer's sign-value corresponds with Shimp and Sharma's (1983) social status dimensions while hedonic value parallels emotional/situational involvement. Imporisk and risk probability seem to be either correlated dimensions or encompassed within the two dimensions suggested by Shimp and Sharma. However, as with other studies, Laurent and Kapferer (1985) did not compare the factor loadings across the 14 products studied.

A final point on the empirical dimensions of existing involvement scales is the rotational methods employed in the exploratory factor analyses used to determine the dimensions. Only Laurent and Kapferer (1985) and Shimp and Sharma (1983) utilized oblique rotations, i.e., allowing dimensions to be intercorrelated. Laurent and Kapferer found interdimensional correlations ranging from .15 to .47 while Shimp and Sharma reported a correlation of .64 between the two dimensions advocated in their study. Furthermore, Shimp and Sharma found the oblique to be superior to the orthogonal rotation in identifying the dimensions in their confirmatory factor analysis. All other scales purporting to measure dimensions of involvement have utilized orthogonal rotations. Although it is plausible that uncorrelated dimensions could be tapping a single underlying construct, it seems more likely that the dimensions would exhibit some degree of intercorrelation, a notion suggestive of oblique factor rotations.

Lastovicka and Gardner's Scale

Lastovicka and Gardner's (1979) involvement scale was selected for examination in the present study for a variety of reasons. First, as suggested by Shimp and Sharma (1983), independent studies are warranted which examine the dimensionality issue across multiple products. The scale tested by Shimp and Sharma was product specific (Bloch's automobile scale). Rather than converting that scale to encompass multiple products, it was deemed appropriate to utilize a scale originally developed to not be product specific. Second, Lastovicka and Gardner's (1979) was one of the earliest multi-item scales developed and has been utilized either conceptually or directly (item usage) in subsequent involvement scales. Third, Lastovicka and Gardner's scale was conceptually based on the foundations of involvement (e.g., Sherif & Cantril 1947). Fourth, the small sample size (n = 40) in Lastovicka and Gardner's study lends itself to criticism given three involvement dimensions based upon 22 items (note, however, that each respondent in their study rated 14 different products resulting in 560 observations). Finally, several of the items in the scale appear to be negligibly related to any of the dimensions (see Table 1), suggesting a more parsimonious solution might be possible via item deletions, oblique factor rotations, or different number of dimensions.

To summarize, using Lastovicka and Gardner's (1979) involvement scale, the present study utilized a large sample in which each participant responded to only a single product out of three potential products. Analysis procedures similar to Shimp and Sharma's (1983) were used. Specifically, exploratory factor results were initially compared to Lastovicka and Gardner's. Using confirmatory factor analysis allowing for both orthogonal and oblique rotations the number of dimensions across products was determined. In addition, correlations between the factor loadings for different products were ascertained. The confirmatory factor analysis results and factor loading correlations were utilized in examining whether involvement was unidimensional, multidimensional, or multidimensional with only a single dimension being consistent across products (e.g., Zaichkowsky 1985; Traylor & Joseph 1984). Given the inconsistent findings in past studies, no specific hypotheses about the empirical dimensions of involvement were made. However, it was predicted the oblique factor rotations would result in a more reliable factor structure than orthogonal rotations (Shimp & Sharma 1983).

METHOD

Sample and Questionnaire. Five hundred sixteen undergraduate students enrolled in arts and sciences, and business classes at a major southern university served as the subjects for this study. As part of a larger investigation, each subject completed a questionnaire containing Lastovicka and Gardner's (1979) involvement scale. Each subject responded to the 22 involvement items for only a single product using a 7-point Likert scale. Three products (shampoo, blue jeans, & athletic/sport shoes) were selected after a thorough review of previous involvement studies and appropriateness for the sample. These products were chosen based on the extent of use in previous involvement research as well as for evidence of consumer involvement variation (cf. Gill & Grossbart 1985). Furthermore, two of the products, athletic shoes and blue jeans, were examined by Lastovicka and Gardner (1979) and were classified as exhibiting different levels of involvement. Two different orders of items were developed. Hence, using a blocked schedule in passing out the questionnaires, subjects responded to one of six different questionnaire forms (3 products, 2 item orders). Incomplete questionnaires, those exhibiting an obvious response set bias, or exceeding a predetermined "floor" score on an embedded lie scale were removed prior to analyses. The final sample consisted of 421 respondents.

TABLE 1

LASTOVICKA AND GARDNER INDICATORS AND LOADINGS WITH OUR EXPLORATORY FACTOR RESULTS

Analysis. Shimp and Sharma's (1983) procedure for testing Block's (1981) automobile involvement scale was, in general, followed here. Similar to Shimp and Sharma (1983), confirmatory factor analysis via LISREL VI was used to provide further evidence of multidimensionality or lack thereof of the Lastovicka and Gardner scale. In brief, Shimp and Sharma's procedure included comparing hypothesized factor models to each other and to the "null" model where complete independence is assumed between all indicators. Fit indices provided by LISREL VI and a calculated index using the null model as a reference (Bentler & Bonett 1980) were used to compare models. Exploratory factor results (i.e., eigenvalues greater than one and scree plots) provided the rationale for comparing additional models beyond the null and the three factor model suggested by Lastovicka and Gardner (see Table 1). Advantages of using Shimp and Sharma's procedure include obtaining an overall fit index of the hypothesized model to the data and determining the extent of interrelations (correlation) between factors.

RESULTS

Exploratory Factor Analysis. Unlike the number of dimensions cited by Lastovicka and Gardner, our exploratory findings suggested a five dimensional solution (see Table 1). These dimensions accounted for 53% of the variation with the first factor accounting for half of the total variance (26.5%). This suggests that involvement, while not unidimensional, may be dominated by one dimension with other dimensions being more peripheral (cf. Shimp & Sharma 1983).

The pattern of loadings is somewhat different from that suggested by Lastovicka and Gardner as well (see Table 1). Factor 1 contains most but not all of the items comprising Lastovicka and Gardner's Normative Importance factor while their first factor was Familiarity. Their Familiarity factor appears to have split into factors 2 and 3. Factor 2 contains items 1, 2, and 3 from Lastovicka and Gardner's Familiarity factor, as well as items 16 and 22 (see Table 1). Factor 3 is composed of three items (4, 5, & 6) reported by Lastovicka and Gardner as also loading on the* Familiarity factor but which now forms a separate, unrelated construct (exploratory results were varimax rotated as per Lastovicka & Gardner). Our exploratory results suggest that Familiarity may be multidimensional: being familiar enough with brand/product features to make comparisons (factor 2) and familiarity as it pertains or leads to preferences (factor 3).

Similarly, factor 4 is composed of two items which formerly loaded on Lastovicka and Gardner's Normative Importance factor. Factor 4 now seems to be an indicator of product importance imparted by external and internal influences. With the exception of items 7 and 10, factor five is composed of items which formerly defined Lastovicka and Gardner's Commitment factor. The pattern of loadings suggests a willingness to switch brands particularly if the consumer receives information which may make them rethink their choice.

In sum, since the exploratory results did not exactly replicate Lastovicka and Gardner's findings, additional, more sophisticated, analyses were warranted to ascertain the dimensionality and item composition of involvement as tapped by the Lastovicka and Gardner scale. Scree plot analysis, for example, suggested that fewer factors might be appropriate. Therefore,-confirmatory factor analysis was used as a further test of involvement dimensionality.

Confirmatory Factor Analyses. The analyses and results from confirmatory factoring are presented in Table 2. A null model was derived postulating no underlying structure and/or a more parsimonious explanation for the data (Bentler & Bonett 1980). In essence, under-the assumptions of the null model, each of the 22 indicators from Lastovicka and Gardner's scale become unitary constructs. The null model represents the "worse case scenario" and served as a subsequent comparison for other hypothesized models.

Other confirmatory analyses compared three, four, and five factor models to each other as well as to the null model for overall fit. Results from exploratory factor models constrained to three, four, and five dimensions provided the rationale for which parameters would be estimated in the confirmatory models. That is, indicator/construct estimation was initially limited to those suggested by the exploratory results (i.e., the lambda y parameter corresponding to the largest absolute loading for each indicator from the exploratory findings was estimated; all other lambda y parameters for that indicator were constrained to zero across constructs). These analyses were performed as an initial step to determine which model(s) seemed appropriate for further investigation. This procedure also enabled a comparison of the more parsimonious (i.e., three factor) solution proposed by Lastovicka and Gardner via constraining parameter estimates to the highest loading for each item according to Lastovicka and Gardner's factor structure (see Table 1). As Table 2 suggests, this factor structure resulted in one of the poorer overall fits and was excluded from further analyses.

Based on initial confirmatory results (see Table 2; models 4 & 5) the three and four factor solutions (again with the lambda y parameters constrained as before) were then reexamined, this time allowing for intercorrelation (oblique rotation) between the dimensions. Finally, using the previously derived fit indices as guides, the three and four factor oblique solutions were then examined allowing all parameters to be estimated to obtain a final best fit between the data and the hypothesized model.

As depicted in Table 2, regardless of constrained versus unconstrained parameters, all models were substantial improvements over the null model. In each case, the LISREL Adjusted Goodness of Fit Index as well as a fit index ratio suggested by Bentler and Bonett (1980) and Shimp and Sharma (1983) indicated considerable improvement when compared to the null model. The five factor solution (Table 2: Model 6, suggested by the exploratory findings) was dropped from further analyses since it did not prove to be an improvement over the three and four factor solutions.

TABLE 2

CHI-SQUARE AND GOODNESS OF FIT VALUES FOR ALTERNATIVE INVOLVEMENT MODELS

Allowing for factor intercorrelations also substantially improved the fit (Table 2: Models 7, 8, 9, & 10) with the best overall fit being the four factor oblique solution. Thus, there appears to be additional evidence that involvement, as measured by this scale, is multidimensional. However, further interpretation of the four factor confirmatory results is warranted. The four factor oblique solution and factor score intercorrelations are presented in Table 3.

With the exception of items 10, 16, & 22 the first factor mimics Lastovicka and Gardner's Normative Importance factor; subsequently labeled "Importance." This factor also appears to be similar to the first factor as per the exploratory results (see Table 1). With two exceptions (item 10 no longer loads, while item 7 now loads on this factor) the fourth factor appears to be similar to Lastovicka and Gardner's Commitment factor and their label was retained. With the addition of item 7, "I usually purchase the same brand within this product class," this factor appears to be tapping brand commitment.

However, as initially suggested by the exploratory findings, Lastovicka and Gardner's Familiarity factor appears to have split (see Table 3; factors 2 & 3). Three of their Familiarity items (1, 2, & 3) have combined with two of Normative Importance indicators (16, 22) to form factor 2, a factor which appears to be familiarity with product/brand features or "Knowledge." Three of their remaining Familiarity items (4, 5, & 6) formed the third factor which has been termed "Brand Preference".

TABLE 3

FOUR FACTOR OBLIQUE SOLUTION LOADINGS AND FACTOR SCORE INTERCORRELATIONS

Although it might be tempting to simply recombine factors 2 and 3 to retain Lastovicka and Gardner's Familiarity factor, the factor intercorrelations (see Table 3) suggest that these constructs, while related (.407), do not contain enough shared variation to be considered essentially similar. Moreover, Table 2 suggests that a four factor solution provides a superior overall fit when contrasted with the three factor solution.

As a final check on the multidimensionality of involvement, exploratory oblique factor analyses restricted to four factors were conducted for each of the three products. Correlation coefficients were then calculated across factors, between the three product categories as well as with the oblique four factor analyses collapsing across products. These correlations were between factor loadings (n = 22 items), not factor scores and therefore, may be indicative of underlying factor structure similarity. These correlations are reported in Table 4.

Factors 2, 3, and 4 for blue jeans and athletic shoes did not appear in the same order as in the collapsed and shampoo factor analyses. These factors were "reordered" for Table 4 to maximize the correlations with the collapsed and shampoo factor loadings. As can be seen in Table 4, factor 1 revealed high factor loading intercorrelations, indicating a reliable factor structure across different products. Furthermore, in all cases, factor 1 accounted for over half of the variance for the four factor solutions.

TABLE 4

CORRELATION COEFFICIENTS BETWEEN OBLIQUE FOUR FACTOR LOADINGS: COLLAPSED ACROSS PRODUCTS AND INDEPENDENTLY FOR BLUE JEANS, SHAMPOO, AND ATHLETIC SHOES

Factor 4 showed acceptable correlations for the three products with the collapsed product factor loadings but, however, not consistently between the three products. Factor 2 showed reasonable correlations between the collapsed, blue jeans, and shampoo but not athletic shoes. Poor correlations were evidenced between the different factor loadings for factor 3 with the exception of shampoo and the collapsed loadings. Only for shampoo and the collapsed factor loadings were consistently high correlations found for all four factors.

CONCLUSIONS

Given the constraints imposed by the scale and products used, confirmatory factor analyses revealed that involvement may be multidimensional when collapsing across products. II terproduct factor loading correlation coefficients suggest that dimensions beyond unity are product specific or, more precisely, potentially product specific. These findings, using a different scale and different methods, lend credence to Zaichkowsky's (1985) and Traylor and Joseph's (1984) use of one reliable dimension across products. The present findings also support and extend Bloch's (1981) and Shimp and Sharma's (1983) position on product specific scales in that dimensions beyond unity may be unreliable across some products. In summary, our results suggest that involvement may be multidimensional both between products and when collapsing across products. However, in utilizing involvement dimensions for a specific products, those dimensions beyond unity should be considered unreliable for subsequent products unless demonstrated otherwise.

Similar to Shimp and Sharma, we found some correspondence between our dimensions and those suggested by Houston and Rothschild (1978), i.e., factors 2, 3, and 4 with the "person specific" aspect and factor 1 with the "social psychological" aspect. However, our dimensions appear to be related to and nested within each other, potentially suggesting the two dimensions delineated by Shimp and Sharma may be multidimensional when multiple products are considered (i.e., items adapted for different products). Obviously this suggests a need for further research and replication which focuses on the dimensionality issue across additional products. Also, although our dimensions closely approximate those found by Lastovicka and Gardner (1979) enough of a difference is noted in our correlated dimensions as opposed to their uncorrelated dimensions to warrant caution in expressing similarity between the dimensions. Finally, our results and Shimp and Sharma's (1983) suggest that the dimensions of involvement are correlated. Future research efforts examining the dimensionality of other involvement scales (e.g., Zaichowsky 1985) are justified.

It would be tempting to postulate that involvement (as a basic construct) entails multiple dimensions that are to some extent product specific. This proposition is unsettling because it would necessitate independent scale development in order to tap involvement with different products. Even though a common dimension(s) might be uncovered, such as importance in the present study, other dimensions may not be congruent and therefore not allow for meaningful comparisons between products or situations. Perhaps, even though between product, situation, or individual differences may exist on factor scores, the underlying factor structure as revealed in factor loadings should be consistent between products and situations for some involvement dimensions. In this study only a single dimension had reliable factor loadings across products. Additional conceptual and empirical involvement research examining dimensionality issues is warranted.

REFERENCES

Bloch, Peter H. (1981), "An Exploration into the Scaling of Consumers's Involvement with a Product Class," in Advances in Consumer Research, Vol. 8, Ed. Kent B. Monroe, Ann Arbor, MI: Association for Consumer Research, 61-65.

Bentler, P. M. and Douglas G. Bonett (1980), ''Significance Tests and Goodness of Fit in the Analysis of Covariance Structures," Psychological Bulletin, 88 (3), 588-606.

Gill, James D. and Sanford Grossbart (1985), "Influence of Detective Claim Strategy and Product Class Evolvement on Beliefs Induced by Deceptive and Corrective Commercials," in Current Issues and Research in Advertising, Eds. James H. Leigh and Claude Martin, Jr., Division of Research, Ann Arbor, MI: Graduate School of Business Administration, 129-160.

Houston, Michael J. and Michael L. Rothschild (1978), "Conceptual and Methodological Perspectives on Involvement," in AMA 1978 Educators' Conference Proceedings, Ed. Subhash C. Jain, Chicago: American Marketing Association, 184-187.

Lastovicka, John L. and David M. Gardner (1979), "Components of Involvement" in Attitude Research Plays for High Stakes, Eds. J. C. Maloney and B. Silverman, Chicago: AMA, 53-73.

Laurent, Gilles and Jean-Noel Kapferer (1985), "Measuring Consumer Involvement Profiles," Journal of Marketing Research, 22 (February), 41-53.

Sherif, Muzafer and Hadley Cantril (1947), The Psychology of Ego Involvement, New York: Wiley.

Shimp, Terence A. and Subhash Sharma (1983), "The Dimensionality of Involvement: A Test of the Automobile Involvement Scale," in Darden, W. R., Monroe, K. B., and Dillon, W. R. (1983), AMA Winter Educators' Conference: Research Methods and Causal Modeling in Marketing, Chicago: American Marketing Association.

Slama, Mark E. and Armen Tashchian (1985), "Conceptualization and Operationalization of Involvement," Journal of Marketing, 49 (Winter). 72-82.

Traylor, Mark B. and W. Benoy Joseph (1984), "Measuring Consumer Involvement in Products: Developing a General Scale," Psychology and Marketing, Vol. 1, No. 2 (Summer) 65-77.

Traylor, Mark B. and W. Benoy Joseph (1985), "Reply to Arora and Baer's Comment on Measuring Consumer Involvement in Products," Psychology and Marketing, Vol. 2, No. 2 (Summer) 127-132.

Zaichkowsky, Judith Lynne (1985b), "Measuring the Involvement Construct," Journal of Consumer Research, 12 (December), 341-352.

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