Redundancy in constructs: Problem, assessment, and an illustrative example

26
J BUSN RES 1991:22:255-280 255 Redundancy in Constructs: Problem, Assessment, and an Illustrative Example Jagdip Singh Case Western Reserve University This article argues that attention to redundancy in constructs is important for theoretical, empirical, and pragmatic reasons. In particular, the article advocates that redundancy between two or more constructs be evaluated from conceptual and empirical perspectives. To facilitate such assessment, this article outlines a systematic, multipronged approach. This approach is based on a directed review of the literature and well-defined methods and principles, such as the Latent Var- iable Structural Equations (LVSE) methodology, and discriminant, convergent, and nomological validity. The author illustrates the suggested approach by utilizing two constructs often used to assess consumers’ attitudes toward business practices- consumer alienation and discontent. The illustration underscores the usefulness of the suggested approach in revealing redundancy gaps that heretofore have remained unidentified. Specifically, consumer discontent and alienation are found to be CCWZ- ceptually nonredundant, but their current operationalizations are not. Implications for future research into redundancy issues are discussed. Introduction The purpose of this article is threefold. 1) The article aims to explicate the nature of the redundancy problem that appears relevant for many social science constructs. I will argue that it is important to address such problems for theoretical, empirical, and pragmatic reasons. 2) The article will outline a systematic, step-by-step ap- proach to evaluate the nature and extent of redundancy in two or more constructs. However, the intent is not to introduce a “new” methodology. Rather, the sug- gested approach is based on well-defined methods and principles, such as the LVSE (Latent Variable Structural Equations) methodology, and the principles of dis- criminant, convergent, and nomological validity. This is particularly advantageous because the use of such well-tested methods provides unequivocal answers to re- dundancy questions. 3) I illustrate the proposed approach by studying the redun- dancy between consumer alienation and discontent constructs. This analysis reveals redundancy gaps that heretofore have remained unidentified. Knowledge of such Address correspondence to Jagdip Singh, Case Western Reserve University, Weatherhead School of Management. Cleveland. OH 44106. Journal of Business Research 22, 255-280 (1991) 0 1991 Elsevier Science Publishing Co.. Inc. 0148-2963/Y l/$3.5(1 655 Avenue of the Americas, New York. NY 10010

Transcript of Redundancy in constructs: Problem, assessment, and an illustrative example

J BUSN RES 1991:22:255-280

255

Redundancy in Constructs: Problem, Assessment, and an Illustrative Example

Jagdip Singh Case Western Reserve University

This article argues that attention to redundancy in constructs is important for theoretical, empirical, and pragmatic reasons. In particular, the article advocates that redundancy between two or more constructs be evaluated from conceptual and empirical perspectives. To facilitate such assessment, this article outlines a systematic, multipronged approach. This approach is based on a directed review of the literature and well-defined methods and principles, such as the Latent Var- iable Structural Equations (LVSE) methodology, and discriminant, convergent, and nomological validity. The author illustrates the suggested approach by utilizing two constructs often used to assess consumers’ attitudes toward business practices- consumer alienation and discontent. The illustration underscores the usefulness of the suggested approach in revealing redundancy gaps that heretofore have remained unidentified. Specifically, consumer discontent and alienation are found to be CCWZ- ceptually nonredundant, but their current operationalizations are not. Implications for future research into redundancy issues are discussed.

Introduction

The purpose of this article is threefold. 1) The article aims to explicate the nature of the redundancy problem that appears relevant for many social science constructs. I will argue that it is important to address such problems for theoretical, empirical, and pragmatic reasons. 2) The article will outline a systematic, step-by-step ap- proach to evaluate the nature and extent of redundancy in two or more constructs. However, the intent is not to introduce a “new” methodology. Rather, the sug- gested approach is based on well-defined methods and principles, such as the LVSE (Latent Variable Structural Equations) methodology, and the principles of dis- criminant, convergent, and nomological validity. This is particularly advantageous because the use of such well-tested methods provides unequivocal answers to re- dundancy questions. 3) I illustrate the proposed approach by studying the redun- dancy between consumer alienation and discontent constructs. This analysis reveals redundancy gaps that heretofore have remained unidentified. Knowledge of such

Address correspondence to Jagdip Singh, Case Western Reserve University, Weatherhead School of Management. Cleveland. OH 44106.

Journal of Business Research 22, 255-280 (1991) 0 1991 Elsevier Science Publishing Co.. Inc.

0148-2963/Y l/$3.5(1

655 Avenue of the Americas, New York. NY 10010

256 J BUSN RES IYY I :22:255-2x0

gaps provides several avenues for future research into the conceptualization and operationalization of the focal constructs. I begin with the explication of the re- dundancy problem.

The Redundancy Problem

The notion of redundancy in two (or more) constructs can be viewed from two perspectives: conceptual and empirical. The conceptual perspective asks if sound theoretical justification is available to view two constructs as logically different conceptualizations. By contrast, the empirical posture questions whether empirical observations (e.g., consumers’ responses) evidence discriminant validity (e.g., by utilizing specific measures) between what researchers may see as two distinct con- structs. Although investigation of such redundancy questions is not common in the literature, researchers in other areas of the social sciences have long argued that lack of attention to redundancy issues results in unnecessary proliferation of con- structs and hinders the process of systematic and cumulative research (Blalock, 1968; Tesser and Kraus, 1976). Thus, for instance, Schwab (1980, p. 25) observes that redundancy in constructs “poses a problem if we take parsimony in scientific explanation seriously.”

Consider, for instance, the measurement of consumers’ attitudes towards busi- ness practices. Interest in this measurement can be traced as far back as the 1940s (e.g., the survey of consumer sentiment by Michigan’s Survey Research Center), and there is every indication that this area will continue to attract the serious attention of researchers (Gaski and Etzel, 1986), practitioners (Harris Poll, 1989) and public policy officials (Wilkie and Gardner, 1974). However, it is only recently that advances have been made in the conceptualization and operationalization of the construct of consumer’s attitude towards the marketplace. Specifically, Lundstrom and Lamont (1976) and Allison (1978) have developed “consumer discontent” and “consumer alienation,” respectively, as two well-defined. psy- chometrically developed constructs.

Unfortunately, when conceptual and empirical redundancy questions are posed for the preceding constructs, disturbingly equivocal answers emerge. For instance, Lundstrom and Lamont (1976) observe that alienation and discontent may be positively related. Lambert (1980) goes as far as to say that these constructs may be “overlapping.” By contrast, Lundstrom et al. (1979, p. 154) concur with Bearden et al. (1983) by noting that discontented consumers “do not express a feeling of alienation” and that the alienated consumers arc “less likely to express their dis- content” (emphasis added). Instances of such unresolved redundancy questions are not uncommon for marketing constructs. A cursory review of the marketing literature revealed unresolved redundancy problems for such “popular” con- structs as role conflict and role ambiguity (Howell et al.. 1989), locus-of-control and inner-, other-directness (Kassarjian, 1965), and power and dependency (Etgar, 1976).

Such unresolved questions pose problems from theoretical, empirical, and prag- matic standpoints. Theoretically. if redundancy issues remain unaddressed, sub- stantial confusion can persist. Some researchers may treat these constructs as distinct and may labor hard to build streams of research focused around each construct; by contrast, others may regard them as “overlapping” and utilize them

Redundancy in Constructs J BUSN RES 1991:22:255-2x0

interchangeably. Clearly this dissipates research energy. Likewise, if empirical redundancy questions go unexamined, our understanding of the current state, sources, and consequences of consumers’ perceptions about business practices will remain suspect, if not misleading. In particular, researchers may attempt to at- tribute, and may, in fact, search for differential sources and/or consequences for, say, the alienation and discontent constructs, when empirical evidence may indicate that consumers perceive no difference between them. Moreover, it is important that (non)-redundancy in conceptualizations correspond to (non)-redundancy in operationalizations. Lack of correspondence raises questions about the measure- ment of the constructs. Finally, from a pragmatic standpoint, it is not useful to have multiple constructs that tap a similar underlying phenomenon without ex- plicitly understanding the nature and extent of the redundancy between the com- peting constructs. Because of these reasons, it is critical that redundancy problems be explicitly investigated. Below I provide a systematic approach for conducting such investigations.

A Systematic Approach to Assessing Redundancy Problems

I propose two sequential procedures for investigating redundancy questions. The first procedure aims at a directed review of the literature, while the second stems from direct empirical analysis. Before the individual approaches are discussed, it is useful to stress the notion of correspondence between the conceptual and empirical evaluations of redundancy. Clearly, if the underlying theoretical framework(s) supports a conceptual distinction between two constructs, such non- redundancies should be reflected in empirical relationships. This aspect will be specifically discussed in the context of nomological validity. For the sake of ex- position, we discuss the proposed approach for two constructs (say, 5, and &). Extension to multiple constructs is relatively straightforward.

Redundancy Evaluation by Review of the Literature

In order to evaluate redundancy issues, I propose initially a review of the literature that is guided by three main objectives. First, it is critical to identify the conceptual definition(s) and the theoretical framework(s) in which the focal constructs are embedded. When this review reveals multiple definitions for a given construct, it may be desirable to evaluate if a particular conceptual definition has gained ac- ceptance. However, in many areas of social sciences, it is likely that neither are sufficiently precise definitions available nor have consensual definitions yet emerged. In such cases, researchers would need to integrate previous research and posit definitions that are precise and at the same time represent the most theoret- ically defensible view of the construct. Once the conceptual definitions of focal constructs are thus identified, they should be evaluated for conceptual redundancy. Specific differences, as well as overlaps (if any), in the conceptualization should be noted and delineated.

Second, differences in antecedents and consequences of the focal constructs should be identified. For the constructs to be nonredundant, identifiable differences must be found in the following relationships: 1) antecedents + 5,) and antecedents --+ &; andlor 2) 5, --, consequences, and 5: -+ consequences. It is not necessary

258 J BUSN RES 1991:22:255-280

J. Singh

that such differences be found for all antecedents and for all consequences of focal constructs. Instead, the necessary (although not sufficient) condition is that at least one antecedent or consequence can be identified for which 5, and & differ in their relationships. Such differences, however, must be substantially and significantly disparate to claim nonredundancy among the focal constructs.

Tesser and Krauss (1976) observe that a “strong” evidence of nonredundancy is obtained when: 1) 5, and & are positively correlated, and 2) 5, and & have significant but opposite relationships (e.g., positive and negative) with a given consequence (or antecedent). This is so because it would be a logical impossibility for 5, and & to be redundant and at the same time be significantly related in an opposite manner to a given antecedent. Furthermore, note that our confidence in this assessment will increase with an increasing number of antecedents and con- sequences for which such strong differences are found.

Third, studies that have utilized both constructs within the same study should be located, and given greater emphasis. Incorporation of both constructs within the same study provides unique opportunities for examining redundancy questions. This is because such studies afford control over such confounding factors as in- strumentation, data collection, and sampling. In this sense, the results from such studies are relatively uncontaminated; thus, such studies should be rigorously ana- lyzed for evidence of redundancy, and possibly given greater weight.

Redundancy Evaluation by Direct Empirical Analyses

Following the review of the literature, researchers may wish to assess empirical redundancy directly between the focal constructs. By direct we imply that the evaluation is not based on studies that specifically did not focus on the redundancy issue. Rather, the study and the analyses are primarily conducted for the purposes of examining the redundancy question. For many constructs, such a direct assess- ment is often not available in the literature. For this reason, it is critical that researchers pay greater attention to designing studies that directly confront the redundancy issue. Below, I outline a three-step approach based on LVSE meth- odology for examining empirical redundancy issues. The first step is aimed at ascertaining redundancy utilizing discriminant and convergent validity principles. The second step evaluates the distinction issue from a nomological validity per- spective. In the last step, the aim is to investigate the specific nature and extent of the overlap between the two measures.

Step I. This step involves formulating a latent variable model for examining the discriminant and convergent validity of the focal constructs. In this model, the underlying latent constructs (5, and &) are posited as reflexive factors of observed fallible indicants. Assume that 5, is measured by p indicants and & by q indicants (model M,, see Fig 1). For multidimensional constructs, p and q may alternatively represent first-order “factors” or dimensions. This is tantamount to using first- order factors (i.e., dimensions) to determine a second-order factor (e.g., aliena- tion). There is precedence for such an approach in both marketing and other areas of social sciences (Dwyer and Oh, 1987). Thus, the latent variable congeneric model can be written as (assuming mean centered data):

Redundancy in Constructs J BUSN RES 1991:22:255-280

259

GDIS GPB UPA PPH PPS

Ic PN T T T T T i T T &I I362 ei3 es4 085

Figure 1. Empirical model for step-l analysis

X=Ae+s (1)

where X is a (p + q) vector of observables or first-order “factors”, 6 is a 2 X 1 vector of theoretical constructs, 6 is a (p + q) X (p + q) vector of unique errors in observables, and A is a (p + q) x 2 matrix of factor loadings. With the assumption of E(e) = E(6) = 0, E(&e’) = 4, and E(6,6’) = T, the variance-covariance matrix of X can be written as

Z5 = A+A’ + + (2)

where C is the variance-covariance matrix of observables, $ is a 2 x 2 matrix of intercorrelations among the theoretic constructs, and JI is a symmetric matrix of error variances in the observables.

The preceding equations can be estimated using a maximum likelihood procedure (e.g., LISREL, Joreskog and Sorbom, 1988). The use of LISREL is desirable because the recovered relationships are between theoretic constructs rather than among some linear combination of observables. Additionally, the obtained solution can be evaluated by several criteria. First, a x2 goodness-of-fit test indicates whether or not the model fits the data. Second, several indicators of the goodness-of-fit (e.g., GFI, RMR, Bentler and Bonnet’s NFI) are available to assess the relative amount of variance-covariances explained by the model. Finally, measurement parameter estimates (i.e., hs in Equation (1)) can be examined for statistical sig- nificance (i.e., corresponding t value > 1.96), and composite reliability can be examined for evidence of significant systematic variance for each 6 (i.e., values 2 0.6; Nunnally, 1967).

To test the null hypothesis that t1 and .$ are nonredundant, initially the inter- correlation between the theoretic constructs (i.e., &,, the off-diagonal term in the + matrix) is examined. For the focal constructs to be distinct, &, ought to be statistically different from 1.0. In other words, the latent constructs should not correlate perfectly if their operational measures are distinct. In addition, for the

260 J BUSN RES 1991:22:255-280

J. Singh

GDIS GPB UPA PPH PPS

Figure 2. Empirical model for step-2 analysis.

null hypothesis to be retained, the latent constructs must satisfy Fornell and Larck- er’s (1981) criterion for discriminant validity. This criterion involves computing, for each 5, the amount of variance extracted as follows:

P”,(E) = ZAi” var(S)

CA,’ var(.$) + 20, (3)

Then, the discriminant validity between 5, and & is achieved if p,,(&) > +22,, for each 5. This is so because pyC(si) represents the proportion of a construct’s variance that is valid. Thus, for a latent construct to be distinct, the variance it shares with any other latent construct (i.e., +22,) should not exceed its own valid variance (i.e.,

P”C(SJ). The alternative hypothesis-that 5, and e2 are, in fact, redundant-is examined

via a test for convergent validity. In particular, an alternative model is estimated that posits that all measures of 5, and c2 converge to a unidimensional theoretic construct. This unidimensional model (say M2) is consistent with the argument that

Redundancy in Constructs J BUSN RES lYY1:22:255-280

261

the operational measures of 5, and & actually tap a single underlying theoretic construct and, thus, evidence empirical redundancy. The statistical test for con- vergent validity is based on comparing two competing LVSE models-namely, model M,, which posits that two distinct latent constructs, 5, and &, underlie the (p + q) measures, and model M,, which presumes a unidimensional factor. The difference in the x2 for the two models (i.e., M,-M,), with the corresponding difference in the degrees of freedom can be used as the test statistic. If the difference statistic is nonsignificant, it suggests that the operational measures for 5, and t2 converge to a single construct and, as such, evidence empirical redundancy (Burnk- rant and Page, 1982).

Step 2. The principles of nomological validity (i.e., measures should relate in a manner consistent with theory; Cronbach and Meehl, 1955) offer another approach to examining redundancy between 5, and &. Specifically, this involves identifying a third latent construct, n, such that 5, and t2 have different theoretical relationships with II. Then, the null hypothesis for the nonredundancy in 5, and E2 is supported if the empirical measures of 5, and e2 also have different relationships with -q, and in the direction expected from theory (Tesser and Kraus, 1976). Conversely, if these relationships between the empirical measures are not different, nonredun- dancy in 5, and c2 is questionable. Furthermore, our confidence in the results (either for or against redundancy) is likely to increase if additional q constructs can be identified that can be subjected to nomological validity analysis.

In terms of estimation, the preceding analysis can be implemented by positing a LVSE model that has one endogenous latent construct (n) and two exogenous latent constructs (k, and e,). The parameters of this model can be estimated using a maximum likelihood procedure (e.g., LISREL; Joreskog and S&born, 1988). The use of this approach is desirable since the estimated coefficients reflect rela- tionships among theoretic constructs rather than among the observable variables themselves or simple linear composites of observables. Also, as discussed earlier, several measures for goodness-of-fit are provided. Distinction between the oper- ational measures is supported when 5, and t2 depict empirical relationships with n that are different and consistent with theoretical expectations. Furthermore, ob- served difference in empirical relationships should not be attributable to sampling fluctuations. Rather, they should stem from systematic differences in the two con- structs. If the preceding conditions are not supported, nonredundancy between the two measures is questionable.

Step 3. If the preceding steps provide support for the redundancy between 5, and t2, then further analysis may be necessary to investigate the extent and nature of overlap. Specifically, statistical tests can be utilized to examine which particular measure(s) contribute most to the redundancy and, conversely, to identify those item(s) that reflect unique aspects of the underlying theoretic constructs. This step can be especially useful in providing guidelines for further refining the operational measures. This involves investigating if one or more measures of 5, load on t2 and vice versa. In terms of estimation, the preceding can be achieved by sequentially freeing up elements of the A matrix to allow for cross-loadings. The t-tests for the elements of A afford a statistical basis for determining which specific indicants cross-load and contribute to the overlap. Note, however, that some minimum

262 J BUSN RES 1991:22:255-2x0

J. Singh

constraints on A are necessary to idenfify-that is, to eliminate the rotational in- determinacy and scaling indeterminacy-the maximum likelihood solution. In par- ticular, with M common factors (in our case M = 2), it is necessary to hold constant at preassigned values, that is, fix M* parameters in the A matrix.

Illustrative Example: Alienation and Discontent Constructs

For illustrating the proposed approach, this study evaluates conceptual and em- pirical redundancy between the consumer alienation and discontent constructs. As noted earlier, redundancy questions remain disturbingly unresolved for the pre- ceding constructs, despite their importance. To the author’s knowledge, no study could be traced that directly examines the relationship between the two constructs. Thus, the current literature concerning the preceding redundancy questions is, at best, equivocal.

As research into the measurement, sources, and consequences of consumers’ attitudes about marketing practices accumulates, it would be helpful to address the preceding redundancy questions explicitly. As such, the illustrative constructs cho- sen for redundancy analysis provide an opportunity to not only demonstrate the proposed procedures but also to address critical substantive issues.

Redundancy Evaluation by Review of the Literature

Appendix I summarizes the previous research concerning the focal constructs. In accord with the focus of this study, Appendix I and the following discussion deal primarily with research involving the specific constructs developed by Lundstrom and Lamont (1976) and Allison (1978).

Consumer Alienation. The theory of alienation has its roots in the writings of Marx, Weber and Durkheim, although contemporary sociologists have significantly refined, expanded, and developed this concept (Lystad, 1972). While early studies evidenced debate about the conceptual status of alienation, more recent studies indicate emerging consensus around four definitional issues. First, researchers be- lieve that while the notion of a “general syndrome” of alienation is useful, a more rewarding approach is to define alienation with respect to a specific object or environmental element (e.g., job, family; see Kanungo, 1979). Consistent with this approach, Allison utilized the “marketplace” as the single, specific environmental focal variable for defining consumer alienation.

Second, the notions of loss of control or feelings of “despair and hopelessness” (Neal and Rettig, 1967) appear central to the concept of alienation. In Seeman’s formulation, these notions are reflected in a multidimensional view of alienation consisting of: 1) powerlessness, a lack of power to influence desirable outcomes; 2) meaninglessness, a separation from meaning in present and future activities; 3) normlessness, an aversion for norms required to achieve goals; 4) isolation, a rejection of common societal objectives; and 5) self-estrangement, a separation from intrinsic gratification. Several researchers attest to the multi-dimensionality of the alienation concept (Neal and Rettig, 1967; Lambert, 1980). Likewise, Allison (1978) defined consumer alienation as separation from the norms and values of market institutions (e.g., manufacturers, retailers), practices (e.g., warranties, re-

Redundancy in Constructs J BUSN RES 1991:22:255-280

turn policies), and from self when one is specifically involved in the consumption role.

Third, researchers have begun to draw clear distinctions between alienating conditions (or antecedents), alienating state, and alienating behaviors. The alien- ation concept is largely associated with a cognitive state of “dissociation” from the marketplace and root necessarily with feelings of satisfaction or dissatisfaction (cf. Kanungo, 1979, p. 129). This is because (dis)-satisfaction is episode-specific while alienation is an enduring state of mind. Thus, it is plausible that an alienated consumer (who is generally dissatisfied) feels satisfied in a particular episode of product/service consumption.

Finally, in terms of alienating conditions, marketing researchers have noted that consumers become alienated as their dissatisfaction with business continues to increase. Structural factors contributing to this condition include, “[t]he rise of self-service retailing, the declining knowledgeability of sales employees, the jux- taposition of the computer between the customer and the organization, and the inherent difficulties of dealing with bureaucracies” (see Aaker and Day, 1982). However, dissatisfaction per se is not sufficient to cause alienation. Instead, situ- ations that constantly frustrate an individual result in alienation because such sit- uations thwart an individual’s need for control. In marketplace situations, continued dissatisfactions that are not satisfactorily redressed in the system could cause al- ienation. In terms of responses, Lambert has suggested that varying amounts of consumer alienation may manifest themselves in somewhat different responses. Low levels of alienation, for instance, may be manifested in expressions of hostility and aggression toward market institutions. In contrast, moderate-to-high feelings of alienation may lead to disengagement behavior, such as not voicing complaints and not seeking redress when dissatisfied. Such consumers, then, are hostile but not aggressive (Bearden, et al., 1983). Furthermore, because such consumers do not express their discontent, they tend to become still more alienated.

Allison (1978) offered a 35-item scale to measure consumer alienation and dem- onstrated that this measure had acceptable reliability (a = 0.86) and construct validity. An empirical study by Bearden and Mason (1983) showed that alienated consumers had: 1) less positive attitudes towards business, 2) less confidence in their ability to evaluate goods and services, 3) less trust in others, 4) less positive attitudes toward complaining, and 5) less satisfaction with their current level of living. In contrast, Clabaugh et al. (1979) found that alienation scores did not differ significantly between satisfied and dissatisfied respondents. In addition, complainers and noncomplainers did not perceive different amounts of alienation. Demograph- ically, alienated consumers have generally been found to be nonwhite and with lower incomes (Allison, 1978).

From a measurement standpoint, the dimensionality of the alienation scale re- mains somewhat ambiguous. Allison (1978, p. 570) concluded that the consumer alienation construct “should be reconceptualized as a unidimensional construct.” However, when Bearden et al. (1983) examined this issue using a split-sample approach, they found support for a 3-dimensional structure. The dimensions were labeled: 1) Business Ethic (BE), including beliefs about the ethical behavior of businesses; 2) Informed Choice (IC), representing opinions about the usefulness of more information and variety in brands; and 3) Personal Norm (PN), which includes statements about one’s personal norms concerning the consumption of

264 J BUSN RES 1991:22:255-2x0 J. Singh

goods and services. Subsequent studies have neither attempted to replicate Bearden et al.‘s results nor examined further the dimensionahty issue.

Consumer Discontent. Unlike alienation, the construct of consumer discontent is not based on a sound theoretical framework. Rather, marketing researchers prompted by the upsurge in the consumer movement during the late 1960s (notably Barksdale and Darden, 1972; Hustad and Pessemier, 1973) proposed the discontent concept in an effort to understand how consumers feel about the business system and to assess “how marketing is doing. 3” More recently, Lundstrom and Lamont (1976) further rehncd this concept based on a comprehensive review of the con- sumer movement.

Specifically, these researchers appear to suggest three central notions concerning the discontent construct. First, discontent is defined as a surrogate measure for consumerism (Lundstrom and Lamont, 1976, p. 373). However, a clear and precise definition of consumerism is lacking in the literature. For instance, Evans (1980) defines consumerism as a “social force within the environment” as well as a wide “range of activities of government, business and independent organizations” di- rccted at protecting the rights of the consumer. This ambiguity in the definition of consumerism defies empirical measurement. Lundstrom and Lamont (1976) appear to circumvent this problem by focusing on the “intensity of consumer dissatisfac- tion” with “market-related practices of business” (p. 373). They specifically in- cluded four types of practices-namely: 1) product strategies, 2) communications and information, 3) the impersonal nature of retailers, and 4) the broader socio- economic forces concerning the business system (e.g., regulation).

Second, Lundstrom and Lamont (1976) posit an attitudinal conceptualization for the discontent concept (cf. p. 374). Although “attitudes” are sometimes defined to include cognitive, affective, and conative components, the discontent concept is most closely associated with the cognitive state of mind or beliefs about business practices.

Third, since discontent is a psychological state, it needs to be distinguished from antecedent conditions and response outcomes. In terms of outcomes, Lundstrom and Lamont (1975) suggest that the consumerist segment (i.e., activist) and the Upset-Action group (i.e., dissatisfied consumers who take some action) tend to report greater levels of consumer discontent. As such, consumers’ discontent is likely to result in overt actions on the part of consumers; people who are discontent (content) with business practices would (not) express aggressive behavior (e.g., voicing complaints) toward market institutions (Hustad and Pessemier, 1973). It is noteworthy that notions of “separation” are not implicated in the definition of the discontent concept.

Although Lundstrom and Lamont do not discuss potential antecedents of dis- content, other researchers have dealt with this aspect extensively. For instance, in their influential book, Aaker and Day (1982) outline nine distinct factors that contribute to consumer discontent. These factors include problem areas such as declining living standards, consumer information gap, and antagonism toward advertising.

Lundstrom and Lamont offered an 82-item measure for the discontent concept, which was shown to have an alpha reliability of 0.96. Subsequent empirical studies report that discontented people are generally found to: 1) be relatively less satisfied

Redundancy in Constructs J BUSN RES 199 1:22:255-280 26.5

with automobiles and footwear (Westbrook, 1980); 2) be more often complainers rather than noncomplainers (Hansen and Macklin, 1984); and 3) favor government intervention in the marketplace (Lundstrom and Lamont, 1976). Demographically, the typical discontented consumer appears to bear a close resemblance to the alienated consumer: married, black, less educated, and the elderly (Lundstrom et al., 1979). In general, empirical work using the discontent measure has suffered due to its length (82 items).

Lundstrom and Lamont have examined the dimensionality of the consumer discontent scale. Using responses from a sample of “general consuming public in a major metropolitan region,” they found support for six interpretable dimen- sions-namely general discontent (GDIS), general pro-business (GPB), inducing unnecessary purchases (IUP), profits and prices too high (PPH), product and pro- motion shortcomings (PPS), and anti-materialistic shift (AMS). Subsequent studies have not investigated these dimensionality issues further. Instead, they have tended to treat the discontent scale as a unidimensional construct.

An Assessment of Redundancy Issues. Are alienation and discontent redundant? From a conceptual standpoint, the preceding review indicates that the focal con- structs are similar in some ways yet substantially different from other perspectives. Both constructs are conceptualized as psychological states representing consumers’ cognitions and beliefs about marketing practices. Furthermore, alienation and dis- content appear to be rooted in consumer dissatisfaction. In other words, a consumer who is generally satisfied with business practices and products/services is most likely neither discontented nor alienated.

However, the notion of loss-of-control or withdrawal is a point of departure between the two constructs. Alienation conceptually implies hopelessness and withdrawal (e.g., separation, helplessness, disengagement) from the alienating object (i.e., marketplace). By contrast, discontent does not evoke any notions of loss-of-control or withdrawal from the marketplace. Rather, Lundstrom and Lamont posit that discontented consumers would most likely be activists and belong to consumerist agencies. In this sense discontented consumers are “en- gaged” or “involved” in the marketplace. This conceptual distinction is consis- tent with the Bearden et al. (1983) observation that alienated consumers are “less likely to express their discontent.” Likewise, the Lundstrom et al. (1979, p. 1.54) contention that, “generally speaking [discontent] consumers do not ex- press a feeling of alienation from the marketplace or the inability to communi- cate their complaints to business,” is successfully explained by this distinction. Thus, while the two constructs share causal factors, they differ remarkably in their structure and response outcomes.

From an empirical standpoint, however, the redundancy question yields an equiv- ocal answer. Allison (1978) notes that the alienation measure is different from discontent because it includes feelings about self in addition to feelings of discon- tent. A typical “self” item from the alienation scale is, “After making a purchase I often find myself wondering ‘why.’ ” Note, however, that the discontent measure also contains similar self-related items; for instance, “The only person who cares about the consumer is the consumer her/himself.” Probably because of this reason, Lambert (1980), upon comparing these measures, concludes that they have a num- ber of common items and thus are “overlapping” measures. Furthermore, empirical

266 _I BUSN RES 1991:22:255-280

J. Singh

studies have produced mixed results. For instance, although alienated as well as discontented consumers are hypothesized to be dissatisfied with life and products, empirical results support this hypothesis more often for discontent (Westbrook, 1980) than for alienation (Clabaugh et al., 1979). Other studies show that discontent relates positively with complaining attitudes and behavior, whereas a less positive or nonsignificant relationship is reported for the alienation measure (Bearden et al., 1983; Lundstrom et al., 1979). In addition, both types of consumer appear to have a similar demographic profile. This finding is surprising because consumerism (i.e., the concept underlying discontent) is generally believed to be a middle-class movement (cf. Table 7.3 in Warland et al., 1986) while alienation is supposedly more common among the “ghetto” consumer (cf. Andreasen, 1986).

In part, this ambiguity in the literature is due to the fact that the alienation and discontent measures have not been utilized in a single study (to the authors’ knowl- edge). This state of affairs obscures our understanding of the direct empirical relationship between the focal constructs. For this reason, the preceding empirical assessment was necessarily based on indirect evidence; that is, drawing inferences from several separate and isolated studies. Another source of confusion derives from the fact that, despite contrary evidence, most studies have tended to treat discontent and alienation as unidimensional measures. The aggregation of multi- dimensional facets possibly obfuscates the assessment of empirical redundancy between the focal measures.

Redundancy Evaluation by Direct Empirical Analyses

Because of the absence of any previous research that directly examined the re- dundancy issue, I designed a study by incorporating both of the focal constructs. In addition, I included a third construct in the study that, from an apriori theoretical standpoint, was expected to relate differently with alienation and discontent. Recall that this allows assessment of the correspondence between conceptual and empirical redundancy. Based on previous studies the multidimensional “attitudes toward complaining” construct was selected; it is hypothesized to have a positive rela- tionship with discontent but a significantly less positive relationship with alienation (Bearden and Mason, 1983; Westbrook, 1980). Below I briefly describe the data and measurement characteristics of the measures before presenting the results from empirical redundancy analysis.

Data. A questionnaire consisting of 82 discontent items, 35 alienation items, and 10 items representing attitude towards complaining was mailed to 1,000 house- holds in the Southwest. As per usual research practice, questions pertaining to a construct were interspersed throughout the instrument in order to avoid systematic effects. A total of 512 responses were received, of which 460 were complete. The usable sample is 65% female, about 42% below 35 years of age, and over 70% with college education. In addition, 85% are white, about 63% are married, and the median income range is $10,001 to $30,000. Comparison of this profile with that obtained from U.S. census data for the geographical area (see Appendix II) reveals quite a few discrepancies, such as the lower percentage of Hispanics in the sample. Readers will note this as a limitation of the data.

Redundancy in Constructs J BUSN RES IYY1:22:255-2x0

267

Measurement Characteristics. Because the specific measurement properties of the focal constructs are not essential to the methodological approach outlined in this article, only brief details are provided here. The analysis was conducted so as to be consistent with previous research. For instance, in the case of the alienation measure, Bearden et al. (1983) have concluded after an extensive assessment that this measure has a 3-dimensional structure. Thus, initially a 3-dimensional structure was examined for the alienation items. Further analyses were performed only if this structure was not supported. This approach was considered desirable because it builds upon previous findings.

Common factor analysis of the 22 alienation items found reliable by Bearden et al. (1983) supported a 3-dimensional structure. The scree plot depicts a leveling off after the third eigenvalue, suggesting a 3-factor solution. The three factors explain 100% of the shared variance in alienation items and about 48% of the total variance. The oblique rotated factor pattern is in Table 1. This factor loading pattern corresponds closely to the pattern obtained by Bearden et al. (1983). Following this study, the alienation dimensions are identified as Business Ethic (BE), In- formed Choice (IC), and Personal Norm (PN). In all cases, a statistically significant loading at @ = 0.05) is obtained for the particular dimension an item was expected to load on (that is, based on Bearden et al.‘s study). For this reason, only such loadings are depicted in Table 1. The interfactor correlations range from 0.27 (IC- PN) to 0.53 (BE-IC).

For the discontent measure, initially a 6-factor solution was obtained by utilizing the specific discontent items (36 total) suggested by Lundstrom and Lamont (1975). The 6-dimensional structure was not supported. The scree plot depicts a leveling off after the fifth eigenvalue, suggesting a 5-factor solution. In addition, none of the items loaded significantly on the sixth factor. Instead, a 5-factor solution appears more plausible. The oblique rotated factor loading pattern is depicted in Table 2. However, to conserve space, only loadings for individual items on their corre- sponding dimension are presented. Although these results are not in agreement with those of Lundstrom and Lamont in regard to the number of factors, the factor- loading pattern depicts a remarkable consistency. Of the six dimensions identified by Lundstrom and Lamont, four are obtained in the present data also with a clear, interpretable structure that corresponds closely to Lundstrom and Lamont’s. These four dimensions are referred to as General Discontent (GDIS), General Pro-Busi- ness (GPB), Profits and Prices Too High (PPH), and Product and Promotion Shortcomings (PPS). The fifth dimension in the present data does not discriminate between the Inducing Unnecessary Purchases (IUP) and the Anti-Materialistic Shift (AMS) dimensions obtained by Lundstrom and Lamont. An examination of the items suggests plausible reasons for this overlap. Both dimensions pertain to “un- necessary” and “wasteful” pressures by business and other people in acquiring material objects. Typical items from the two dimensions underscore this possibility. For instance, for IUP the item with the highest loading in Lundstrom and Lamont’s data is: “Companies encourage the consumer to buy more than he really needs.” The corresponding item (i.e., highest loading) for AMS is: “People rate other people by the value of their possessions.” For the purposes of this study, the fifth dimension is referred to as Unnecessary Purchases and Anti-materialism (UPA). The interfactor correlations range from insignificant (GPB-PPS) to 0.50 (GDIS- PPH).

268 J BUSN RES 1991:22:255-280

Table 1. Factor Coefficients for Alienation Items

Coefficient Factor/Item (alpha reliability)

Business ethic (BE) [alpha = 0.841

0.61 It is not unusual to find out that businesses lied to the public.

0.56 Stores do not care why people buy their products just as long as they make a profit

0.65 Business’s prime objective is to make money rather than satisfy the customer.

0.37 It is difficult to identify with business practices today.

0.56 Unethical practices are widespread throughout business.

0.49 Products are designed to wear out long before they should.

0.46 A product will usually break down as soon as the warranty is up.

0.60 Business is responsible for unnecessarily depleting our natural resources.

0.44 Most claims for product quality are true.”

0.52 Harmful characteristics of a product are often kept from the consumer.

0.42 Advertisements usually present a true picture of the product.”

0.44 Most companies are responsive to the demands of the consumer.”

Informed choice (IC) [alpha = 0.651

0.76

0.44

Even with so much advertising it is difficult to know what brand is best.

The wide variety of competing products makes intelligent buying decisions more

difficult.

0.27 It is hard to understand why some brands are twice as expensive as others.

0.61 It is difficult to know what store has the best buy.

0.38 It is difficult to identify with current trends and fads in fashion.

Personal norm (PN) [alpha = 0.621

0.58 After making a purchase I often find myself wondering “why.”

0.46 I often feel guilty for buying so many unnecessary products.

0.54 1 tend to spend more than I shculd just to impress my friends with how much I have.

0.32 One must be willing to tolerate poor service from most stores.

0.48 Buying beyond one’s means is justifiable through the use of credit.

“Denotes items that are rcversc scored.

The attitudes toward complaining construct is measured using 10 items based on Richins (1982). Eigenvalues of the reduced correlation matrix (i.e., scree plot) suggest a 3-dimensional structure. In all? the three eigenvalues explain 100% of the shared variance, and 55% of the total variance. When the factors are rotated using oblique rotation, a clear interpretable pattern of loadings is obtained (see Table 3). In addition, this pattern corresponds closely to the structure reported by Richins (1982). These dimensions are labeled as: 1) societal benefits (SO), the extent to which consumers believe that complaining would result in societal benefits, 2) norms of complaining as related to self (SE), i.e., to what extent is complaining an “appropriate” behavior for oneself, and 3) norms pertaining to other people

Redundancy in Constructs

Table 2. Factor Coefficients for Discontent Items

J BUSN RES 1991:22:255-280

269

Coefficient Factor/Item (alpha reliability)

General discontent (GDIS) [alpha = 0.851

0.31 Warranties would not be necessary if the manufacturer made the product right in the first place.

0.59

0.30

Products that last a long time are a thing of the past.

Food that is not nutritious is another example of business trying to make a buck and not caring about the customer.

0.44 When the product is advertised as “new” or “improved ” it is the same old thing only in a

different package.

0.27 0.42

0.40

Many times the salesman says one thing to the shopper but he knows it’s just the opposite. The only person who cares for the consumer is the consumer himself.

Business is the one using up our natural resources but it does nothing to replace what has been taken.

0.53 Many companies listen to consumer complaints but do nothing about them.

0.47 Products fall apart before they have had much use.

0.67 Companies aren’t willing to listen or do anything about consumer gripes.

0.68 Business does not help local residents because it is not profitable.

0.59 The consumer is usually the least important consideration to most companies.

General pro-business (GPB) [alpha = 0.661

0.35 Salesmen really take an interest in the consumer and make sure he finds what he wants.’ 0.41 Business firms usually stand behind their products and guarantees.” 0.40 The quality of goods has consistently improved over the years.”

0.30 The actual product I buy is usually the same as advertised.”

0.42 Business takes a real interest in the environment and is trying to improve it.” 0.43 Generally speaking, products work as good as they look.”

0.36 Companies are helping minorities and the under-privileged by providing them with jobs.”

0.43 The business community is actively involved in solving social problems.”

0.50 The information on most packages is enough to make a good decision.”

0.30 Most products are safe when they are used right.”

Profits and prices too high (PPH) [alpha = 0.631

0.35 Business profits are too high.

0.27 The Government should enforce ethical business standards.

0.36 Business profits are high yet they keep on raising their prices.

Product and promotion shortcomings (PPS) [alpha = 0.571

0.45

0.53 0.46

Styles change so rapidly a person can’t afford to keep up.

Advertised “specials” aren’t usually in the store when the shopper goes there.

A warranty or guarantee may be a good one but the service department is often unable to do the work correctly.

Unnecessary purchases and anti-materialism (UPA) [alpha = 0.751

0.56 Stores advertise special deals just to get the shopper into the store to buy something else.

0.58 All business really wants to do is make the most money it can.

0.64 Companies encourage the consumer to buy more than he really needs.

0.50 Salesmen are pushy just so they can make a sale.

0.33 Companies try to influence the government just to better themselves.

0.30 Credit makes things too easy to buy.

0.39 People rate other people by the value of their possessions.

0.36 Business has commercialized many meaningful holidays, such as Christmas.

“Denotes items that are reverse-scored

270 J BUSN RES 1991:22:255-280

J. Singh

Table 3. Factor Coefficients for Attitudes Toward Complaining Items

Coefficient Factor/Item (alpha reliability)

Societal benefits (SO) [alpha = 0.671

0.61

0.62

0.30

By making complaints about unsatisfactory products. in the long run the quality of

products will improve.

By making complaints about defective products. I may prevent other consumers from

experiencing the same problem.

Complaining isn’t much fun. but it has to be done to keep business from becoming

irresponsible.

Norms pertaining to self (SE) [alpha = 0.60]

0.73

0.62

0.43

0.52

It bothers me quite a hit if I don’t complain about an unsatisfactory product.

It sometimes feels good to get my dissatisfaction and frustration with the product off

my chest by complaining.

1 often complain when I am dissatistied because I feel it is my duty to do so.

I feel a sense of accomplishment when I manage to get a complaint to a store taken

care of.

Norms pertaining to others (OT) [alpha = 0.55 1

0.53 People are bound to end up with unsatisfactory products once in a while, so they

should not complain.”

0.25 People have a responsibility to tell store when a product they purchase is dcfcctive.

0.78 I don’t like people who make complaints to stores because usually their complaints are

unreasonable.”

(OT), i.e., the extent to which complaining is perceived to be desirable behavior for other people. The interfactor correlations range form 0.24 (OT-SO) to 0.38 (SE-SO).

Empirical Redundancy Results

Step I. The LVSE model of Figure 1 was formulated with three alienation and five discontent dimensions as measures of their corresponding theoretic constructs. This model as well as other competing models were estimated by submitting the covariance matrix to the maximum likelihood procedure in LISREL VII. Table 4 summarizes the results obtained. In overall fit, model M, (i.e., corresponding to Fig 1) produced the following statistics: x2 = 132, df = 19, RMR = 0.065, GFI = 0.90. Although the xz statistic suggests that the covariance matrix reproduced does not equal the observed covariances, readers will note that the problems with the x2 test in general, and for problems involving relatively large samples (i.e., n = 460) in particular, are well documented (Fornell and Larcker, 1981). Thus, in evaluating the model, other indicators of fit are given greater prominence.

In particular, note that compared with the null model of no underlying constructs (M,,), model M, shows a significant improvement with a nonnormed-fit-index (NFI) of 0.912 (Bentler and Bonnet, 1980). Consistent with this, the GFI and RMR for this model also approach acceptable values. In addition, all t-values for the param- eters of A matrix are statistically significant. Finally, composite reliabilities for the discontent ([,) and alienation (&,) latent constructs are 0.85 and 0.81. respectively

Redundancy in Constructs

Table 4. Results from Empirical Redundancy Analysis”

J BUSN RES 1991:22:255-280

271

Model Description X2 df GFI RMR

Model tests

MC, Null modelb

M, Two-construct model (see Fig. 1)

M, One-construct model

Model comparisons

WI-M, Test for the fit of the two-construct

model

&-M, Test for the fit of the one-construct

model

M,-M, Test for the convergent validity of

the two constructs

1928.5 28 0.37 0.34

132.3 19 0.90 0.06

133.2 20 0.90 0.06

1796.2 9 0.912d

1795.3 8 0.917d

0.9”’ 1 0.004d

Test for dixriminant validity (based on model M,, Fig. 1)

Composite reliabilities’

Alienation 0.81

Discontent 0.85

Intercorrelation

$21 1.02

Variance extracted (P,,(.$,))~

Alienation 0.61

Discontent 0.55

“All models were estimated by the method of maximum likelihood using LISREL VII.

“The null model is based on the premise that the alienation and discontent measures have zero intercorrelations

and, thus, there are no underlying theoretical factors.

‘This is the difference between the x2 values for the models under comparison.

‘%is is the non-normed fit index based on Bentler and Bonnet (1980). ‘This is nonsignificant at p = 0.05. ‘The computational formula for composite reliability is given in Fornell and Larcker (1981).

8The computational formula for variance extracted is given in Fornell and Larcker (1981).

(see Table 4). Based on Nunnally’s (1967) criterion, the constructs appear to be reasonably measured with significant nonrandom variance.

In terms of empirical redundancy, note that in Table 4 the standardized cor- relation between alienation and discontent latent constructs is estimated to be slightly over 1 .O. This implies that 5, and & are perfectly correlated, thereby ques- tioning the notion that the two constructs are empirically distinct. Furthermore, application of Fornell and Larcker’s (1981) criterion also calls into question the distinction between the focal measures. The variance extracted by discontent (5,) and alienation (&) constructs is estimated (by using Equation (3)) as 0.55 and 0.61, respectively (see bottom panel of Table 4). In an absolute sense, the amount of variance extracted by the individual constructs appears respectable. However, when these values are compared with the variance shared between the 5s (+22, = 1.04), Fornell and Larcker’s (1981) criterion for discriminant validity is not satisfied. In other words, the two constructs share more variance in comparison with the “true” construct variance that they individually extract.

Furthermore, to check if these results were robust, step 1 analyses were per- formed by utilizing all of the 82 discontent and the 35 alienation items. First, the discontent and alienation items were individually factor analyzed, and then the factor scores were submitted to LVSE analysis -as per Figurk 1. Similar results were obtained (x’ = 194, df = 19, GFI = 0.82, and RMR = 0.104). In particular,

272 J BUSN RES 1991:22:255-280

Table 5. Nomological Validity Results for Empirical Redundancy Analysis“

J. Singh

First-Order Factors Second Order Factor

Parameter” q = so q = SE q = OT (Fig. 2)

Factor loadings

GDIS 0.76 0.76 0.75’ 0.76

GPB 0.25 (0.04) 0.25 (0.04) 0.25 (0.04) 0.25 (0.04)

UPA 0.59 (0.04) 0.60 (0.04) 0.60 (0.04) 0.60 (0.04)

PPH 0.71 (0.06) 0.71 (0.06) 0.72 (0.06) 0.72 (0.06)

PPS 0.73 (0.05) 0.73 (0.05) 0.73 (0.05) 0.73 (0.05)

BE 0.67 0.67 0.67 0.67

IC 0.55 (0.05) 0.54 (0.05) 0.54 (0.05) 0.54 (0.05)

PN 0.45 (0.06) 0.45 (0.06) 0.47 (0.06) 0.45 (0.06)

SO 0.93 - 0.45’

SE 0.93 0.79 (0.30)

OT 0.89 0.30 (0.12)

Structural coefficients

Discontent + q 0.20 (0.16) 0.27 (0.15) 0.29 (0.16) 0.28 (0.09)

Alienation + 7 0.03 (0.17) 0.09 (0.16) 0.17 (0.17) 0.12 (0.09)

“All modeis were estimated by the method of maximum likelihood using LISREL VII. hAll parameter estimates are standardized values with SE in parentheses. ‘This parameter was fixed to unity to fix the vale of measurement. As such, SE for this parameter was not

available.

&, was estimated at 1.12, and the variance extracted did not exceed variance shared. This suggests that results are robust across reduced and full scales.

The evidence from convergent validity also calls into question the premise for nonredundancy between the alienation and discontent measures. Initially, model M?, which posits a unidimensional model, was estimated. This model yielded the following fit statistics: x2 = 133, df = 20, RMR = 0.065, GFI = 0.90. Compared with model M, (i.e., the 2-factor model), model MZ is statistically nonsignificant (x’ difference = 0.90, df = 1, p > 0.0s). This suggests that the hypothesis that discontent and alienation measures converge to a single factor cannot be rejected. In summary, step 1 analysis overwhelmingly suggests that discontent and alienation constructs are not empirically distinct.

Step 2. Just as the alienation and discontent constructs were estimated from their first-order dimensions, the attitude construct was estimated utilizing its three first-order dimensions (SE, SO, and OT). This model is displayed in Figure 2, and the maximum likelihood parameter estimates are in the last column of Table 5. To supplement this analysis, 3 additional models were estimated by utilizing each

of the attitude dimensions individually. That is, the nomological validity analysis was performed at the second-order factor level (i.e., Fig 2), as well as the first- order factor level (i.e., by using SE, SO, and OT individually). Readers will note that there is some overlap among the first-order factors of attitudes toward com- plaining (because of intercorrelations). However, this overlap is not troublesome, since less than 15% variance is shared between any two first-order factors of the attitudes construct. For this reason, the results from the first-order analysis can be evaluated rather independently. The results from the first-order analysis are in the first three columns of Table 5.

In terms of the overall fit, the second-order model of Figure 2 yielded the

Redundancy in Constructs J BUSN RES 1991:22:2ss-280

273

following statistics: x2 = 287, df = 41, GFI = 0.90; RMR = 0.06. For the first- order models, the x2 ranges from 216 to 138 with df = 25, GFI 2 0.90, and RMR 5 0.05. While the x2 statistic is significant for all models, the GFI and RMR values indicate that a significant portion of the observed covariances are explained by the posited model. In addition, note that all measurement parameters (i.e., factor loadings) are statistically significant (i.e., in Table 5 all estimates are at least twice as large as their standard errors) and relatively stable (i.e., in Table 5 factor loadings for alienation and discontent vary by less than 5% across the four models) indicating that the measures are specified correctly.

More importantly, Table 5 indicates that although the nomological relationships are in the expected direction, they evidence redundancy. For instance, in the second-order model, discontent has a positive direct effect of 0.28 on attitude, as expected. In contrast, alienation has a relatively less positive direct effect, 0.12, also as hypothesized. However, the 95% confidence intervals suggest that the preceding effects are overlapping. Specifically for the second-order model, the 95% confidence intervals are (0.05 to 0.53) and { -0.12 to 0.40) for discontent and alienation, respectively, indicating a significant overlap (Sawyer and Peter, 1983). Furthermore, this hypothesis for the overlap in nomological effects is upheld for each of the 3 first-order models. For instance, with SO as the dependent construct, the respective 95% confidence intervals are { -0.11 to 0.51) and { -0.30 to 0.36). Consistent with this, when SE is the dependent con;iruct, the corresponding in- tervals are { - 0.03 to 0.57) and { - 0.23 to 0.41). This consistent overlap in effects rejects the notion that the discontent and alienation measures are able to reproduce the theoretically expected differences in their nomological relationships. Never- theless, readers should note the dependence of the preceding result on a single multidimensional external construct as a limitation of the study. Research with additional antecedents and consequences is necessary for stronger conclusions.

Step 3. Because the redundancy between alienation and discontent appears ten- able, this step specifically investigated which particular dimension(s) contribute most to the redundanzy. As noted earlier, this can help provide guidelines for future research. Several models were estimated, guided by the modification indices for the model in Figure 1. The model finally selected allowed cross-loadings for the GPB and PPH dimensions of the discontent, and the BE dimension of alien- ation. All coefficients for the A matrix were significant at p = 0.05 (t values range from 1.9 to 21). In terms of overall fit, this model is superior to the congeneric model (x2 diff = 33, df = 3, p < 0.01).

In this final model, the only discontent dimension that did not appear to load on alienation was PPS, i.e., product and promotion shortcomings. Perhaps, general discontent (GPB), feelings that prices and profits are too high (PPH), and general anti-materialism (UPA) tendencies are greater sources of alienation than is PPS. For alienation, only the business ethic (BE) dimension had a significant cross- loading on discontent. Somewhat surprisingly, the informed choice (IC) and the personal norm (PN) dimensions appeared to measure alienation unambiguously. Apparently the feelings of self (PN) with respect to the consumption role are more closely tied into alienation, as speculated by Allison (1978). These are, however, post hoc explanations of results uncovered from an exploratory procedure. In addition, even this final model indicated high modification indices for some elements

274 J BUSN RES 1991:22:255-280

J. Singh

of the 86 matrix, indicating that more commonalities between discontent and alienation were present. Freeing these elements resulted in problems of instability and identification. For these reasons, caution is warranted in drawing inferences from the above results.

Discussion and Directions for Future Research

This study set out to: 1) highlight the redundancy problem, 2) provide a general and systematic approach for assessing such problems, and 3) illustrate the suggested approach. 1 have argued that determining the redundancy among two or more constructs is critical for theoretical, empirical, and pragmatic reasons. Theoretically, unrecognized redundancy undermines scientific progress and accumulation of re- search. Empirically, when not explicitly documented, redundancy poses serious doubts on our understanding of the phenomena and on our ability to provide useful guidelines to interested constituencies. Pragmatically, redundancy is innately trou- blesome. For these reasons, the current neglect of redundancy issues in social sciences is unfortunate and warrants serious attention.

To address such issues, I have argued that redundancy should be assessed from conceptual and empirical perspectives. The conceptual perspective evaluates if sound theoretical arguments can be garnered to posit that the focal constructs are conceptually different. The empirical redundancy question, however, is posed in the context of estimated relationships between measures of the focal constructs. Because it is plausible that constructs may be conceptually nonredundant while their measures fail to achieve nonredundancy (and vice versa), lack of correspon- dence between conceptual and empirical evaluations presents serious problems.

In addition, this article has provided a systematic and multifaceted approach for investigating conceptual and empirical redundancy questions. First, it is recommend that previous research be reviewed so as to: 1) identify (or develop) clear, precise, and consensual definitions of the focal constructs; 2) identify one or more ante- cedents and/or consequences for which the focal constructs evidence and/or are hypothesized to possess disparate relationships; and 3) locate and closely analyze previous research (if any) that incorporates the focal constructs within the same study. Nonredundancy between the focal constructs is supported if this analysis reveals clear differences in the definition, causes, and effects of the focal constructs.

However, because redundancy issues have received inadequate attention, it is likely that unequivocal insights will not be forthcoming from the literature. Con- sequently, so as to confront redundancy issues directly, I propose a 3-step approach. The first two steps are directed at determining if redundancy between the focal measures is tenable. By contrast, the third step attempts to isolate the specific dimensions (or items) that contribute to the redundancy (if it exists). Furthermore, the analyses suggested in these steps are based on well-defined methods and prin- ciples. This approach, therefore, can be utilized with reasonable assurance of its validity.

Is the proposed approach effective in understanding redundancy between dis- content and alienation? The literature review revealed that sound theoretical jus- tification was forthcoming to posit that the discontent and alienation constructs are logically different conceptualizations. This conceptual nonredundancy stems from the consideration of loss-of-control notions. Loss-of-control or withdrawal is central

Redundancy in Constructs J BUSN RES 1991:22:255-2x0

275

to the definition of alienation. By contrast, discontent as a surrogate for consum- erism implies absence of withdrawal cognitions. Instead, discontent is more closely associated with an “engaged” or “involved” stance toward the marketplace. This conceptual difference appears to underlie the disparate behavioral responses of alienated (i.e., to not complain when dissatisfied) and discontent (i.e., to be a consumer activist) consumers. Nevertheless, a conceptual relationship between the two constructs was also supported. ‘l’his relationship stems from common causal factors that are rooted in consumer dissatisfaction. Thus, we appear to have suf- ficient theoretical evidence to claim that alienation and discontent are nonredun- dant, though related, constructs.

With regard to the empirical redundancy issue, previous research has been equiv- ocal. In contrast, our findings qre clear, consistent, and compelling. Each of the first two steps led to the conclusion that the empirical distinction between alienation and discontent measures is not tenable. In particular, step I analysis suggested that the measures lack discriminant validity and, instead, appear to achieve convergent valid- ity. In step 2, the redundancy was examined by investigating the nomological rela- tionships with the attitudes construct. No significant difference was found in the empirically estimated nomological relationships in any of the four models estimated. This consistency in the results, I believe, is compelling.

In addition, analysis conducted in step 3 provided an insight into the nature of this redundancy. Specifically, with the exception of PPS, all dimensions of discon- tent and the BE dimension of alienation appeared to contribute significantly to the redundancy. An examination of the BE items indicates that while these items measure dissatisfaction they do not include loss-of-control notions (typical item: “Products are designed to wear out long before they should.“). This probably explains why this dimension is so closely related to discontent. By contrast, alien- ation items relating to meaninglessness (PN items) and informed choice (IC) did not cross-load and produced unique factors, thereby suggesting little contribution to the redundancy. Readers will note that these alienation items tap notions of loss-of-control (typical item: “After making a purchase I often find myself won- dering “why?“). This suggests that the major cause for the empirical redundancy in the current measures may be that these measures do not faithfully tap the conceptual distinction between the focal constructs. Additionally, the uniqueness of PN and IC items indicates that it may be possible to posit other operationali- zations that are not empirically redundant by utilizing the conceptual basis for nonredundancy (i.e., loss-of-control). Caution is advised in the interpretation of these results, however, because of the exploratory nature of these analyses. Fur- thermore, the limitations of the results due to underrepresentation of Hispanics, the cross-sectional design, and the geographical restriction of the sample should be noted. Future replications with different samples and in different contexts would be needed to understand the generalizability of the findings.

The findings, however, should not be taken to imply that the important scale development work of Allison or Lundstrom and Lamont ought to be discarded. Iustead, this article is a call for further refining of these scales. The 82-item dis- content and the 35-item alienation measures provide a useful starting point for a scale refinement process (cf. Churchill, 1979), leading to measures that more faith- fully reproduce the underlying conceptual distinction between them. Of course, this process may also involve generating new and more appropriate items. I believe

276 .I BUSN RES 19!491:22:255-280

that this line of research should precede further investigation into the sources and effects of these constructs. Clearly, antecedents and consequences of a phenomenon are more meaningful after the phenomenon itself has been conceptually and em- pirically mapped without ambiguity.

This study is also a call for examining general redundancy problems. Because much scale development work in social sciences is conducted without an explicit investigation of its overlap with other constructs, it is suspected that the redundancy among constructs may be a more prevalent problem than heretofore recognized. Two recent articles underscore this point. In the first, Morrow (1983) examined the concept of “commitment to work” and found that researchers, over time, had developed in excess of 25 different commitment-related concepts and measures. Based largely on a conceptual analysis, Morrow (1983, p. 486) noted that the various concepts and measures are “partially redundant and insufficiently distinct to warrant continued separation.” Thus, she concludes that “more rigorous con- struct validation” studies are critically needed to examine redundancy issues. This study has offered a methodological framework to examine these issues precisely. In the second instance, Qualls (1987) found that the construct of “sex-role ori- entation” had three different measures: 1) the Bems Sex Role Inventory, 2) the Osmond and Martin Sex Role Attitude Scale, and 3) the Scanzoni Sex Role Scale. Based on an empirical analysis of these measures, Quails (1983) concluded that although “each measure purports to capture the same construct,” the various measures are not tapping the underlying construct “in the same manner and, thus, lack convergent validity.” I have argued that such lack of correspondence is a by- product of the inattention to redundancy issues in the literature.

Can such redundancy problems be avoided in future research? The answer is yes. Utilizing the proposed approach, I have demonstrated that an explicit assess- ment of the conceptual and empirical redundancy among constructs is tractable. In future research, two approaches are recommended. First, researchers must consciously examine other constructs that are conceptually similar to the focal construct in their studies. Second, and perhaps more important, redundancy issues should be routinely investigated in future scale development work. I hope that this study will provide the necessary impetus for such work.

References

Aaker, D., and Day, G., A Guide to Consumerism, in Consumerism: Search for the Con- sumer Interest. D. Aaker and G. Day, eds., Free Press, New York, 1982.

Allison, N. K., A Psychometric Development of a Test for Consumer Alienation from the Marketplace. Journal of Marketing Research, 15 (November 1978): 565-575.

Andreasen, A., Disadvantaged Consumers in the 198Os, in The Future of Consumerism, P. Bloom and R. Smith, eds., Lexington, CA, 1986.

Barksdale, H. C., and Darden, W. R., Consumer Attitudes Toward Marketing and Con- sumerism, Journal of Marketing, 36 (October 1972): 28-35.

Bearden. W. O., and Mason, J. B., Empirical Evidence on Marketplace Alienation, Journal of Macro Marketing, (Fall 1983): 6-20.

Bearden, W. O., Lichtenstein, D. R., and Teel, J. E., Reassessment of the Dimensionality, Internal Consistency and Validity of the Consumer Alienation Scale, American Edu- cators Conference Proceedings (1983): 35-40.

Bearden, W. O., Mason, J. B., and Teel. J. E.. An Investigation of the Relationships among

Redundancy in Constructs J BUSN RES 19!41:22:255-280

277

Education, Income Adequacy, Consumer Alienation and Life Satisfaction, in Znterna- tional Fare in Consumer Satisfaction and Complaining Behavior, vol. 7. Ralph Day and Keith Hunt, eds., 1983, pp. 26-34.

Bentler, P. M., and Bonnet, D. G., Significance Tests and Goodness of Fit in the Analysis of Covariance Structures, Psychological Bulletin, 88 (1980): 588-606.

Blalock, H. M., Jr., The Measurement Problem: A Gap Between the Languages of Theory and Research, in Methodology in Social Research, H. M. Blalock and A. B. Blalock eds., 1968, pp. 5-27.

Burnkrant, R., and Page, T. Jr., An Examination of the Convergent, Discriminant, and Predictive Validity of Fishbein’s Behavioral Intention Model, Journal of Marketing Research, 14 (November 1982): 550-561.

Churchill Jr., Gilbert, A., A Paradigm for Developing Better Measures of Marketing Con- structs, Journal of Marketing Research, 16 (February 1979): 64-73.

Clabaugh, Jr., Maurice, G., Mason, J. B., and Bearden, W. O., Consumer Alienation and Causal Attribution as Moderators of Consumer Satisfaction/Dissatisfaction and Com- plaint Behavior, in New Dimensions of Consumer Satisfaction and Complaining Behav- ior, vol. 3. Ralph Day and Keith Hunt, eds., 1979, pp. 2-9.

Cronbach, J. L., and Meehl, P. E., Construct Validity in Psychological Tests, Psychological Bulfetin, 52 (1955): 281-302.

Darden, W., Carlson, S. M., and Hampton, R., Issues in Fitting Theoretical and Mea- surement Models in Marketing, Journal of Business Research, 12 (1984): 273-296.

Dwyer, R. F., and Oh, S., Output Sector Munificence Effects on the Internal Political Economy of Marketing Channels, Journal of Marketing Research, 24 (November 1987): 347-358.

Etgar, M., Channel Domination and Countervailing Power in Distributive Channels, Journal of Marketing Research, 13 (August 1976): 254-262.

Evans, J., A New Approach to the Study of Consumerism, in Consumerism in the United States, J. Evans, ed., Praeger, New York, 1980.

Fornell, C., and Larcker D., Evaluating Structural Equation Models with Unobservable Variables with Measurement Error, Journal of Marketing Research, 18 (February 1981): 39-50.

Gaski, J., and Etzel M., The Index of Consumer Sentiment Toward Marketing, Journal of Marketing, 50 (July 1986): 71-81.

Hanson, R. A. and Macklin C., Criterion-Related Validity of the Lundstrom and Lamont Consumer Discontent Measure: An Extension of the Validation Process, American Marketing Association Proceedings (1984): 60-63.

Harris P., 54% percent of All Adults Feel Alienated and Powerless, The Harris Poll Press Release, Jan. 1, 1989.

Howell, R., Wilcox, J., Bellenger, D., and Chonko, L., An Assessment of the Role Conflict and Role Ambiguity Scales, in Eficiency and Effectiveness in Marketing, G. Frazier et al., eds., 1989, pp. 314-319.

Hustad, T. P., and Pessemier, E. A., Will the Real Consumer Activist Please Stand Up: An Examination of Consumers’ Opinions about Marketing Practices, Journal of Mar- keting Research, 10 (August 1973): 318-324.

Joreskog, K. G., and S&born D., LISREL VII: A Guide to the Program and Applications, Chicago: SPSS, Inc., 1988.

Kanungo, R. The Concepts of Alienation and Involvement Revisited, Psychological Bulletin. 86 (1979): 119-138.

Kassarjian, H., Social Character and Differential Preference for Mass Communication, Journal of Marketing Research, 2 (May 1965): 146-153.

Lambert, Z. V., Profiling Demographic Characteristics of Alienated Consumers, Journal of Business Research, 49 (1981): 23-40.

278 J BUSN RES 1991:22:255-280 J. Singh

Lambert, Z. V., Consumer Alienation, General Dissatisfaction and Consumerism Issues, Journal of Retailing, 56 (Summer 1980): 3-24.

Lambert, Z., and Kniffin F., Consumer Discontent: A Social Perspective, California Man- agement Review, 18 (Fall, 1975): 36-44.

Lundstrom, W., and Lamont, L., The Underlying Dimensions of Consumer Discontent: A Factor Analytic Study of the Consumer Discontent Scale, in Combined Proceedings, Edward Mazze, ed., 1975, pp. 238-245.

Lundstrom, W. J., and Lamont, L. M., The Development of a Scale to Measure Consumer Discontent, Journal of Marketing Research, 13 (November 1976): 373-381.

Lundstrom, W. J., Skelly, G., and Sciglimpagila, D., How Deep are the Roots of Consumer Discontent: A Study of Rural Consumers, in New Dimensions of Consumer Satisfaction and Complaining Behavior. Ralph Day and Keith Hunt, eds., 1979, pp. 153-156.

Lystad, M., Social Alienation: A Review of the Current Literature, The Sociological Quar- terly, 13 (Winter 1972): 90-113.

Morrow, Paula C., Concept Redundancy in Organizational Research: The Case of Work Commitment, Academy of Management Review, 8 (1983): 486-500.

Neal, A., and Rettig S., On the Multi-Dimensionality of Alienation, American Sociological Review, 32 (February 1967): 54-64.

Nunnally, J. C., Psychometric Theory: McGraw Hill, New York, 1967.

Quails, W. J., Household Decision Behavior: The Impact of Husbands’ and Wives’ Sex Role Orientation, Journal of Consumer Research, 14 (September 1987): 264-279.

Richins, M. L., An Investigation of Consumers’ Attitudes Toward Complaining, Association of Consumer Research Proceedings, vol. 9. A. Mitchell, ed., 1982, pp. 502-506.

Sawyer, A., and Peter, P. J., The Significance of Statistical Significance Tests in Marketing Research, Journal of Marketing Research, 20 (May 1983): 122-134.

Schwab, D. P., Construct Validity in Organizational Behavior, Reseurch in Organizational Behavior, 2 (1980): 3-43.

Seeman, M., On the Meaning of Alienation, American Sociological Review, 24 (1959): 783- 791.

Tesser, A., and Krauss H., On Validating a Relationship Between Constructs, Educational and Psychological Measurement, 36 (1976): 111-122.

Warland, R., Herrmann R., and Moore D., Consumer Activism, Community Activism, and the Consumer Movement, in The Future of Consumerism, P. Bloom and R. Smith, eds., Lexington, CA, 1986.

Westbrook, R. A., Interpersonal Affective Influences on Consumer Satisfaction with Prod- ucts, Journal of Consumer Research. 7 (June 1980): 49-54.

Wilkie, W., and Gardner D., The Role of Marketing Research in Public Policy Decision Making, Journal of Marketing, 38 (January 1974): 38-47.

Redundancy in Constructs J BUSN RES 1!491:22:255-280

279

280 J BUSN RES 1991:22:255-280 J. Singh

Appendix II

Demographic Characteristics of the Sample”

Characteristic Sample U.S. Census

Sex Male

Female

Ageh

5 2.5 yrs

26-35 yrs

36-45 yrs

46-55 yrs

z 56 yrs

Education

< High school High school

College

Graduate

Income

5 $10,000

$lO,OOl-30.000 $30,001-50.000

$50,001-70.000

> $70,000

Ethnic White

Black

Hispanic

Other

Marital status Married

Divorced Separated

Widowed Single

35 50

65 50

12 22

30 26

20 15

13 24

25 24

3 19 26 41

55 29

16 11

I7 30

47 58 25 7”

7 3

4 2’

85 72

5 7

8 20

2 1

63 56

7 6 I3 2

7 8 10 ?P

“All values are in percentages. ‘For the census information. population units of 20 years or more in age were considered ar . .._ .argct popular~u,,

of interest. ‘U.S. Census did not provide comparable values for this group. Thus, this percentage represents the income group

$lO,oolL34,999. “Similarly, this value actually pertains to mcome group $35,WJ-49,999. ‘Pertains to U.S. Census value for the income group $50.001-74.999. ‘Pertains to income group > $75,lXKl.