Predicting Consumer Choice Probabilites By Causal Models of Competition

Volker Trommsdorff, Technische Universitat Berlin
ABSTRACT - The article has four objectives. The first is to develop method for competition assessment in product image research. The second is a methodological contribution to improve the acceptability of market research: the interactive process of exploratory and confirmatory competition analysis is outlined and demonstrated. As a third objective the linear structural causal analysis is transferred to competition research. The fourth objective is to show how response tendencies in image rating scales can be controlled within the LISREL methodology.
[ to cite ]:
Volker Trommsdorff (1984) ,"Predicting Consumer Choice Probabilites By Causal Models of Competition", in NA - Advances in Consumer Research Volume 11, eds. Thomas C. Kinnear, Provo, UT : Association for Consumer Research, Pages: 601-606.

Advances in Consumer Research Volume 11, 1984      Pages 601-606

PREDICTING CONSUMER CHOICE PROBABILITES BY CAUSAL MODELS OF COMPETITION

Volker Trommsdorff, Technische Universitat Berlin

[I want to express my thanks to Dr. Lutz Hildebrandt, Science Center Berlin, and to Jurgen Tauchnitz, Technische Universitat Berlin, for computing many more LISREL models than could be presented here. Thanks also to the unnamed company cooperating in that project.]

ABSTRACT -

The article has four objectives. The first is to develop method for competition assessment in product image research. The second is a methodological contribution to improve the acceptability of market research: the interactive process of exploratory and confirmatory competition analysis is outlined and demonstrated. As a third objective the linear structural causal analysis is transferred to competition research. The fourth objective is to show how response tendencies in image rating scales can be controlled within the LISREL methodology.

1. Competition as a marketing research concept

Relative concepts such as price differences, market shares, relative market positions, competitive advantages etc. are most prominent in the marketing practice. Nevertheless in the attitudinal marketing research literature relations like these are seldom mentioned. The main reason may be that models of attitude measurement have been transferred to marketing research without regard to the relativity of brand attitudes. The original social psychological theories and models were designed to investigate a single attitude object, not its relations to competing objects. The typical attitude objects were one s job, minorities, habits (Fishbein & Ajzen 1975). Fishbein himself supported transferring the model into the marketing area (Ajzen and Fishbein 1980, pp. 149-172) but did not create an adjusted concept of relative or competitive attitudes. Of course it is possible to assess attitudes toward single brands and to consider their competitive relations in subsequent stages of analysis. But this procedure will be shown not to reflect the reality of consumer behavior and the requirements of the marketing management. If marketing researchers had applied social psychological models more critically they likely would dispose of better instruments of competition analYsis by now.

Of course practical mar'<et research considers competitive relations, at least qualitatively. Several theoretical approaches analyzing competition are well known. For instance the early german marketing theory, based on neoclassical microeconomics, used coefficients of cross-price-elasticity as indicators of competition intensity (Gutenberg 1955, 246 ff.). Within the actual methodology of marketing research brand positioning models can be use to express competitive relations. Distances between the brand positions are interpreted as intensities of competition, explicitly proposed e.g. in Urban s (1975) model PERCEPTOR. Special German contributions to methodology of competition analysis were given by Dichtl, Andritzky and Schobert (1977), Kaas (1977), and Mazanec (1976). They al follow the principle of decomposing global brand evaluations, preferences, or similarities to deduce indicators of competitional aspects between these brands.

Dichtl et.al. developed a multivariate procedure to differentiate the "relevant market", a key concept of German competition regulations. A matrix of interbrand similarity coefficients is used to decide whether a brand is competing, i.e. belongs to the relevant market. The similarities are calculated not from objective or subjectively perceived attributes but from consumption situation characteristics expressed as frequency distributions of consuming each brand in different situations, that is to satisfy different needs. The similarity matrix is decomposed into competitive product attribute dimensions by Nonmetric Multidimensional Scaling. The remaining brand space can be analyzed by Cluster Analysis to reveal relevant markets as clusters of brands which are highly similar, i.e. competitive. Kaas was interested in measuring market share response functions on price differences of competing brand pairs. Degrees of competition are indicated by elasticity coefficients of these functions. There are two possibilities to get input data. First, retail panel data of market shares and actual price differences. Second, consumer preference data on pairs of brands including some amounts of money to represent prices within the usual price range of the product. The preference data were analyzed by pairwise comparison scaling following Thurstone's law of comparative judgement, case V. The preference shares are treated like market shares. Both procedures lead from overall brand values to indicators of competition. Mazanec proposes to ask consumers directly for perceived similarities between competing pairs of brands to assess substitutional competition.

Different from these examples is the recent approach of Laroche and Brisoux (1981). They assume a direct influence of attitudes toward competing brands (including the "own" brand) on buying intentions. Common to the aforementioned German suggestions they treat competition from a global brand evaluation view. Because the model to be proposed here is attached to the author's multiple brand attitude determination of buying intentions, the procedure will be described in section 2.

Neither of these methods and models are adequate to analyze the generation of competitive relations, that is consumer s impression formation of brands and attributes. Competition cannot only be established by existing attitudes, preferences, and over-all similarities but also by momentary perceptions of separate brand advantages and disadvantages, by processing single information chunks or advertisements of unique selling propositions.

The common attitude theory-oriented measurement models of marketing research cannot cope with such direct competitive effects because they are restricted to elaborate attribute structures. The usual kind of model typically defines intensity of competition as an interbrand distance in an attribute space equivalent to a similarity coefficient of attribute profiles. Contrary to this concept there are kinds of real competition where no comparability of profiles and no common attribute space is existing in consumer s perceptions. In addition often distinct attitudes do not even exist especially in low involvement product classes. The general lack of research in nonrational consumer behavior has recently been shown by Holbrook and Hirschman (1982). At least the consequence for market research methodology is to impeach the concepts of similarity and distance as outstanding indicators of competition.

The divergence between model and reality does not only concern the completeness of impressions of competitive brands but also the process of impression formation. Common models, at least implicitly, assume rational information processing of all relevant product attributes just like scoring models operate. More psychological processing as recently systematically described by Bleicker (1983) cannot be consistent with-this kind of competition analysis.

A realistic model to explain competition related consumer behavior should be more flexible. First it has to represent real consumer information processing strategies, even the most simplistic ones. Second competitive effects should be implemented at realistic stages Of the impression formation process, not only referring to existing attitudes. Even fleeting glance impressions of single cues can be decisive and therefore should be incorporated in the model if necessary. Third the model should allow for incomparable attribute profiles between brands - without a detour to attitude formation. In sum the traditional perspective of multi-brand-multi-attribute information matrices will be weakened considerably. Some formal accuracy is to be sacrificed for the sake of pragmatic validity.

So far the discussion deals with the individual consumer behavior level. But to use competition research data for marketing decisions is to use aggregate information from markets and segments. Different individuals may have different evoked sets of brands, different attitude structures and different brand attribute perceptions as well as interbrand advantage/disadvantage perceptions. To switch from individual to aggregate units of research implies some difficulties. But it is beyond the scope of this contribution to discuss details of these problems. In principal our suggestions can be extended to market segments, if the requirement of homogeneity is met sufficiently.

2. Laroche and Brisoux's multibrand attitude model

The approach is based on a thought of Woodside and Clokey (1974). Their aim was to categorize the main multiattribute attitude models in using the following four criteria:

- independent variables: beliefs only / beliefs and evaluations / beliefs and importances

- aggregation across dimensions: yes / no

- if yes, how to aggregate: multiplicative / additive

- consideration of brands: only one / several

Following the last mentioned criterium the authors define > multibrand type of attitude models. All possible attribute values of all brands are added to an overall predictor of intention to buy a specific brand. The model is logically wrong and of no practical use. However its value gas heuristic in stimulating LAROCHE and BRISOUX (1981) to develop a new multibrand model which seems to overcome the most serious deficiencies.

In a short form the model appears as:

I = M A

being

I = vector of intentions to buy brand j

A = vector of attitudes toward brand J

M = matrice of parameters to regress intention to buy brand j on all attitudes toward brands, including brand j.

In comparison to that model the traditional monobrand model is represented in the diagonal of M. The model requires to assume that the consumer retrieves a set of existing attitudes toward all relevant brands to form a choice. The dependent variable (buying intention) measured by constant sum rating scales can be treated as buying probability or market share on the aggregate level of analysis. The independent variables (brand attitudes) are measured by simple rating scales. The regression parameters separately estimated for each brand j indicate degrees of influence of brand attitudes on intentions to buy brand j- Values within the diagonal of M represent the well-known attitude-intention relations of monobrand models. The off-diagonal values represent the influence of liking or disliking other brands and can therefore be interpreted as overall interbrand competition indicators.

In testing the model empirically the following hypotheses were confirmed:

- each buying intention is best predicted by the attitude toward that brand, diagonal values of M excel the off-diagonal values

- attitudes toward other brands (off-diagonal values) contribute significantly to the explanation of buying intentions

- attitude effects of competing brands are always signed negative

- matrice M is not symmetrically, that is the opposite off-diagonal values jk and kj differ from each other. An attitude toward one brand may affect the intention toward another one but the reverse may not be true.

On average the model explained 54.6% of the buying intention variance. The improvement to 51.2% when only attitudes toward the "own brand" were used is not too much. With data of a similar type of investigation (see 4.: empirical example) the present author achieved 18% versus 11% explained variance (see table 2). The high explanatory power level of Laroche and Brisoux s investigation is probably due to response set in the data. The model should not only be judged by the explained variance criterion. The authors themselves mention three critical aspects. First, the model may be incomplete because only attitudes are considered as predictors of intentions. Second, interaction effects between concurrent attitudes are neglected by the procedure of linear regression analysis. Third, homogeneity of the sample may not always be true particular in differentiated markets.

3. Structure of a flexible model of competition

For the sake of constructive criticism the main problems should be specified. The regression methodology needs independent predictors (Cohen and Cohen 1975, p. 116). Contrary to that condition, attitudes toward competing brands usually are correlated at least caused by subject-scale-interaction effects. This kind of artifact will probably be increased using single raw rating scales instead of elaborated measurement models. Some kind of scaling and data adjustment procedure is indicated.

To cope with remaining interbrand correlations more sophisticated procedures are available instead of regression analysis. Also the first-mentioned global self criticism Of omitting other predictors than attitudes can be specified: A more flexible model should overcome the restriction on perfect brand attitudes. Less stable brand perceptions and less rational information processing should be reflected by the model if necessary.

If it is sufficient to explain the intention to buy brand A by means of the total attitude toward brand A, then only this predictor should be included into the model. If one or more attitudes toward competitive brands substantially contribute to the explanation they have to be considered in addition. Special attention has to be paid if no attitude exists but only more or less cursory impressions. That may be a single product attribute belief, perhaps from a unique selling proposition, a general impression about quality, an information chunk like company image, geographical origin, price class or personal reference. If those impressions affect the intention to buy brand A they should be included as predictors.

An intention to buy has to be explained by at least one single impression. The pattern of predictors may change between brands to be predicted. To regress each brand on different predictors is as possible as to use one common predictor for all brands. A full attribute by brand information matrice as the Pattern of predictors for all brands is an "ideal" case.

In summary, the following types of predictors may be used:

- a global attitude value

- multiattribute attitude structures

- single brand attributes

- single information chunks of brand A and/or of some selective brands or even of all competing brands.

On an individual level the number of buying intention causes normally is small, increasing with the product involvement level. As mentioned it is not intended to explain individual but segmental buying behavior. Since segment homogeneity means that all consumers are guided by the same causes the number of effective predictors in the model increases with segment heterogeneity. Therefore it is suitable to begin the analysis with a model of many brands and attributes and to reduce the number of predictors successively. The final model should reflect the number of buying intention causes to be counted in a market segment across individuals. An index of fit will help to find an appropriate number of predictors.

Aside empirical efficiency, the relevance for marketing management is a criterion to select an appropriate sample of brands, attributes, and brand-by-attribute cells. The reason is that some empirically effective brand attributes cannot or will not be influenced by marketing policy, e.g. by reasons of ethics, law, costs. Contrarily some less effective predictors may be incorporated in the model because of long range marketing strategy or of strong influence through actual marketing decisions.

4. Exploratory and confirmatory methodology

Both aspects, empirical efficiency as well as managerial relevance lead to the methodological core of the flexible competition model. The distinction between exploration and confirmation, quite similar to Reichenbach's distinction of the context of discovery and the context of justification, has been neglected in marketing research. Exploratory research is aimed at the generation of marketing information, e.g. listing the names of competitors or measuring market shares. Confirmatory research is aimed at the test of present hypotheses. The decision is either to maintain or to reject a competition hypothesis, e.g. that a specific advantage of a competing brand affects the buying intentions of brand j. Of course, everyday market research has exploratory and confirmatory purposes. But the researcher is not always aware of them.

The problem is not to classify each research question, for instance on a scale between extreme exploratory and extreme confirmatory research. The problem is rather that market research practice does not always handle the different purposes appropriately, partly as a result of a one-sided view of social research methodology. The relevant literature has mainly been concerned with problems in the context of justification (Guttman 1977). Methods designed for the context of discovery are rare to find. The same impression is reflected in the literature of applied statistical methods. Apart from exceptions like Tukey's (1977) Exploratory Data Analysis the methods of exploration seem to be no theme for statisticians. Fully neglected was the fact that exploratory and confirmatory methods are contingent on each other because both research aspects are mutually related in the process of theory construction.

On the other hand mainly descriptive functions were attributed to everyday market research (Aaker and Day 1980, p. 53). Especially the usual image and attitude research to describe competition among brands is said to be exploratory in nature (Ajzen and Fishbein 1980, p. 158): Relevant attributes nave to be collected and measured. In contrast to an exploratory concept the management has more or less concrete hypotheses about relevant dimensions of competition, both as regards from the empirical aspect of consumer behavior and from the managerial aspect of marketing decisions.

Therefore Deshpande and Zaltman (1982) are right to attribute problems of market research acceptance to the interaction between the-researcher and the manager: the research briefing does not state adequately whether to explore or to confirm. Usually the management is not likely to tell the hypotheses the researcher. Nevertheless the results are expected to confirm the hypotheses. Otherwise the research (instead of the hypotheses) will probably be rejected, adding to an unfavorable image of market research.

If an investigation can only be partially based on theoretical knowledge, an interactive process of exploration and confirmation is called for. This type of research is often regarded as unscientific because of misunderstanding scientific methodology. Especially the principle of falsification is often stressed one-sided and puristic, increasing the gap between expectations and performance of market research.

The model proposed requires an awareness of the need to define exploratory and confirmatory targets and steps of competition research in interaction between management and researcher. As a first confirmatory step we require the management information input. That is not only the set of brands considered to be competitive but also their hypotheses about general and brand specific competition attributes. In the language of consumer information processing research the required information consists of rows, columns, and single cells of an information display matrix. Brands considered competitive appear as rows, attributes as columns, and single brand advantages or disadvantages (unique selling propositions) as cells. In contrast to consumer information processing research the matrix may contain several empty cells.

Usually we cannot decide whether to reject or confirm the proposed competition effects by a single step empirical test but by several interactive steps of specifying, test and respecifying the model. Estimated power of explaining buying intentions, either as single path coefficients or as total model efficiency coefficient, serve as decision criteria in that process. The structure of the final model is than gained by both, confirmatory input from management and exploratory input from data analysis.

From experience we know that the number of single competition factors in a final model is rather small. Even if the number of initially proposed brands and attributes define information matrices up to about a hundred cells the interactive modelling of competition effects will reveal only a handful of final effects usually caused by three to five competitors and by three to five attributes relevant for at least one brand. In case of a product market successfully differentiated by emotional conditioning (Kroeber-Riel 1980, 126 ff), possibly only one global evaluative attribute (attitude, if any) across all relevant brands is effective. Then, as a special case of the model proposed, the model equals the Laroche and Brisoux type . Another special case is that of competition by means of only one specific product attribute, e.s. price.

In principle the measurement models to operationalize competitional attributes are free. To analyze markets with distinct image competition (the beer market, for instance). it is proposed to apply a standard image questionnaire, for instance the "Imagedifferential" (Trommsdorff 1975). This is a marketing research version of Osgood's Semantic Differential. At best an integrated measurement- and substantial causal modelling methodology should be applied as a tool of data analysis . For a simultaneous test of validity each attribute should be assessed by at least two indicators. To apply the concept of concurrent validity the indicators Of a common attribute should be different in measurement logic, for instance generated first by questioning and second by content analysis of sales agent reports.

the LISREL linear structural causal analysis methodology (Joreskog and Sorbom 1978, Bagozzi 1980, Hildebrandt 1983) is the most sophisticated procedure to analyze competition Dy the proposed model, hitherto (Hildebrandt and Trommsdorff 1983). More simple models without considering ,measurement error and interdependent causal effects can be performed by multiple regression analysis. Further development in multivariate statistics may also allow the Specification of nonmetric causal competition models.

5. Research example

to demonstrate the model proposed we start with a multiple regression model followed by a multistage LISREL analysis. The data were taken from a marketing research survey sponsored by a major German brewery. Beside scientific purposes the company was interested in image and product positioning data of their regional beer market. The sample of 500 was representative of the general target group of peer consumers in that region. A commercial institute carried out the field research using our questionnaire summarized in table 1.

the predictor variables used in the models discussed below are absolute differences between ideal and real image atTributes, indicating evaluative aspects of each attribute. the attributes were selected together with the management to use their prior knowledge about image competition in the market, confirmatory in the sense proposed above. Two to four items were selected to represent each of the more or less connotative denotative attributes (quality, tartness, rooted in ones native land. exclusiveness).

Rating data are contaminated by several errors and response sets (Saal, Downey and Lahey 1980). Leaving out One social desirability set the most serious errors in product image ratings are: extreme (leniency) response Sets and halo effects (Beckwith, Kassarjian and Lehmann 1978). To analyze a set of rating data by competition effects they should be preadjusted or statistically controlled. The true variance should not be affected seriously. Some part of extreme response set will already be reduced by taking the difference scores between ideal and real ratings. To adjust the data completely by the individual style of using rating scales a z-standardization by rows (not by columns as usual done in linear statistic programs) is recommended (Trommsdorff 1977) especially if only regression analysis without control of measurement error will be applied. If the measurement error is retained for statistical control within the model, factors like extreme response set and halo effect can be modelled as separate causal paths or intercorrelations permitted between competition predictors. An example is given below (fig. 4).

TABLE 1

QUESTIONNAIRE SUMMARY

To compute a basic model without consideration of measurement error by regression analysis the following steps of data adjustment, selection, and transformation were performed:

- standardization by rows to reduce response error

- exploratory factor analysis and factor score computation to obtain orthogonal attributes

- exclusion of runaway values defined as the 5% upper and lower standardized scores

- selection of the special target group of nonloyal beer consumers

- if real values can exceed ideal, linear transformation of ideal-real difference scores to obtain equal regression weights below and above the ideal point, according to the part worth function model (Green and Srinivasan 1978. 106).

Then each regression was carried out. For instance consider a model to explain the intention to buy brand A (quasi metric constant sum scale). Four competitors 8, C, D, E, and the four attributes mentioned above were selected. Within the (4+1)x4 cells "information matrix", only 15 predictors were specified as competitive initially (to depict the model see circles and arrows in fig. 3). The regression model equals the LISREL model. without specification of measurement error and construct intercorrelation. Eight of 15 predictors were revealed to De significant, accounting for 52% of the variance of the dependent variable. Compared with alternative models (table 2) the proposed model s power is outstanding.

TABLE 2

EXPLANATION POWER COMPARISON OF ALTERNATIVE MODELS

The second model (see fig. 3) is an extension Of the usual regression model. The Model is aimed to cope with a kind of systematic error to be reckoned with rating profile data. The effect formerly called subject-scale-interaction will be referred to as scale effect. To model scale effects in terms of the LISREL methodology a specific factor for each attribute could be specified as an additional cause of all brand ratings on that attribute. Then the model equals a two step confirmatory factor analysis. A less complicate approach is to allow for correlation between brands within attributes. But to estimate the model also require that we specify a linear structural equation model. To evaluate a LISREL model the criterion "chi square / degrees of freedom" is applied as shown by Schmitt and Bdeian (1982). The model can be accepted if the index is between 1 and 10. Hence the second model is accepted (486/80=6.1).

FIGURE 3

INTERMEDIATE MODEL: TOO MANY FACTORS, MEASUREMENT ERROR UNSPECIFIED, ONLY SCALE EFFECTS SPECIFIED

Beyond response effects specific to scales or attributes, brand effects may occur: Attributes within brands may be correlated systematically. The common term referring to this effect in rating scales is halo (Huber and James 1978). As discussed in the case of scale effects, halo could be specified as specific factors influencing the ratings of all attributes of each brand. Judd and Krosnick (1982) used a similar specification in another research context. Alternatively halo here is assessed as free interscale correlation across attributes.

As a consequence of the proposed interactive process of exploratory and confirmatory data analysis together with the management, the final model (fig. 4) is now restricted to only six relevant competition factors. The model does not only consider scale and brand effects by respective correlations to be estimated. To assess each factor two adjusted ideal-real-differences on items indicating that factor we-e specified confirmatory. The remaining error of each of these measurement models can be seen from values of each second path because the first path is restricted to unity for technical purposes. The parameter values of arrows directed to n indicate the strength of influence of the remaining competition factor on intention to buy brand A.

The model can be accepted with a chi square value of 265 and 52 degrees of freedom (the fit index beeing 5.1). The model summarizes the proposed interactive research process, achieving decomposition of usual image research data into two distinct and one general type of measurement error on one hand and into six competition factors on the other hand. The company controls three, and two competitors together also control three competition factors. The structure and the quantities of effects did not surprise the management but gave them confirmatory advice for strategic marketing decisions beeing discussed at that time. As expected the first attribute (general impression of quality) has some influence concerning the "own" brand A as well as the competing brands B and E. According to the first hypothesis of Laroche and Brisoux evaluation of brand A predicts better than of others.

FIGURE 4

FINAL MODEL, REDUCED BY ATTRIBUTES AND BRANDS, CONSIDERING RESPONSE, SCALE AND BRAND EFFECTS

More specifically the management expected to compete sharply with competitor E on the second attribute (tartness). The model confirms that hypothesis. Brand A s (outstandingly high) tartness is the most powerful cause of intentions to buy or not to buy brand A. But the scale effect between brands A and E is extraordinary high. That is, in evaluating both brands on tartness there is almost no difference (corresponding to advertising claims and real tartness values). Because the scale effect is calculated from only two brands its interpretation as a systematic error cannot be trusted here. The value of -.2 from E 8 to n means that a slight decrease of intention to buy Brand A follows from an outstanding tartness image of Brand E. The model connections of E 12 indicate that exclusiveness is no real competition factor. After partializing halo effects the influence from an exclusive Brand A image on the corresponding buying intention is quite small and does not compete with that of other brands.

Increasing the sophistication of market research models as shown has its price. The additional expenditure in theoretical, analytical, and computational work is respectable. Further research effort is required until a general model of causal analysis of competition can be offered.

REFERENCES

Aaker, David A. and George S. Day (1980), Marketing Research: Private and Public Sector Decisions, New York: Wiley and Sons.

Ajzen, Icek and Martin Fishbein (1980), Understanding Attitudes and Predicting Social Behavior, Englewood Cliffs, N.J.: Prentice-Hall.

Bagozzi, Richard P. (1980), Causal Models in Marketing, New York: Wiley and Sons.

Beckwith, Neil E., Harold Kassarjian, and Donald R. Lehmann (1978), "Halo Effects in Marketing Research: Review and Prognosis," Advances in Consumer Research, Vol.5, ed. H. Keith Hunt, Ann Arbor, MI: Association for Consumer Research, 465-457.

Bleicker, Ulrike (1983), Produktbeurteilung der Konsumenten, Wurzburg: Physiica.

Cohen, Jacob and Patricia Cohen (1975), Applied Multiple Regression/Correlation Analysis for the Behavioral Science, Hillsdale, N.J.: Lawrence Erlbaum Ass.

Deshpande, Rohit and Gerald Zaltman (1982), "Factors Affecting the Use of Market Research Information: A Path Analysis," Journal of Marketing Research, 19, 14-31.

Dichtl, Erwin, Klaus Andritzky, and Rudolf Schobert (1977), "Ein Verfahren zur Abgrenzung des relevanten Marktes auf der Basis von Produktperzeptionen und Praferenzurteilen", Wirtschaftswissenschaftliches Studium, 6, 290-301.

Fishbein, Martin, Ajzen, Icek (1975), Belief, Attitude, Intention and Behavior, Reading, Mass. a.o.: Addison-Wesley

Gutenberg, Erich (1955), Grundlasen der Betriebswirtschaftslehre. 3d.2: Der Absatz, Berlin: Springer.

Guttman, Louis (1977), "What is What in Theory Construction," Discussion Paper No. 28, Berlin: Science Center.

Hildebrandt, Lutz and Volker Trommsdorff (1983), "Konfirmatorische Analysen in der empirischen Forschung", in Innovatorische Marktforschung, ed. Werner Kroeber-Riel et.al., Wurzburg: Physika. 139-160.

Hildebrandt, Lutz (1983), Konfirmatorische Analysen on Modellen des Konsumentenverhaltens, Berlin: Duncker & Humblot.

Holbrook, Morris B. and Elizabeth C. Hirschman (1982), "The Experimental Aspects of Consumption: Consumer Fantasies, Feelings, and Fun," Journal of Consumer Research, 9, 132-140.

Huber, Joel and William James (1978), "A Measure of Halo," Advances in Consumer Research, Vol. 5, ed. H. Keith Hunt, Ann Arbor, MI: Association for Consumer Research, 468-473.

Joreskog, Karl G. and Dag Sorbom (1978), LISREL - Users Guide, Uppsala: University of Uppsala, Department of Statistics.

Judd, Charles M. and Jon A. Krosnick (1982), "Attitude Centrality, Organization and Measurement," Journal of Personality and Social Psychology, 42, 436-447.

Kaas, Klaus P. (1977), Empirische Preisabsatzfunktionen bei Konsungutern, Berlin: Springer.

Kroeber-Riel, Werner (1980), Konsumentenvernalten, 2nd ed., Munchen : Vahlen

Laroche, Michel and Jaques E. Brisoux (1981), "A Test of Competitive Effects in the Relationship Among Attitudes and Intentions," Working Paper, Montreal: University o Montreal.

Mazanec, Josef (1976), "BMDIC - Ein demoskopischer Indikator zur Messung der Intensitat der Substitutionskonkurrenz zwischen Produktmarken," Arbeitspapier der absatzwirtschaftlichen Institute der Wirtschaftsuniversitat Wien Nr. 5/1976.

Saal, Frank E., Downey, Ronald G. and Mary Ann Lahey (1980), "Rating the Ratings: Assessing the Psychometric Quality of Rating Data," Psychological Bulletin, 88, 413-428.

Schmitt, Neal and Bdeian, Arthur G. (1982)," A Comparison of Two LISREL and Two-Stage Least Squares Analysis of a High Hypothesis - Life Job Satisfaction Reciprocal Relationship," Journal of A>plied Psychology, 67, 806-817.

Trommsdorff, Volker ( 1975) , Die Messung von Produktimages fur das Marketing, Koln: Carl Heymanns.

Trommsdorff, Volker (1977), "Probleme der Image- und Einstellungsmessung im Werbepretest," in Personal und Sozialorientierung der BWL. Bd.3: Marketing, ed. Jerbard Reber, Stuttgart: Poeschel, 61-84.

Tukey, John W. (1977), Exploratory Data Analysis, Reading, Mass.: Addison-Wesley.

Urban, Glen L. (1975) , "PERCEPTOR: A Model for Product Positioning," Management Science, 21, 858-871.

Woodside, Arch G. and Clokey, James D. (19?4), "Multi-Attribute/Multi-Brand Models," Journal of Advertising Research, 14(5), 33-40.

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