Development and Testing of a Cross-Culturally Valid Instrument: Food-Related Life Style

Karen Bruns°, The Aarhus School of Business
Klaus G. Grunert, The Aarhus School of Business
ABSTRACT - Based on a cognitive perspective, we propose to make life style specific to certain areas of consumption. The specific area of consumption studied here is food, resulting in a concept of food-related life style. We have developed an instrument that can measure food-related life style in a cross-culturally valid way. To this end we have developed a pool of 202 items, collected data in three countries, and have constructed scales based on cross-culturally stable factor patterns. We have then applied the set of scales to a fourth country, in order to further test the cross-cultural validity of the instrument.
[ to cite ]:
Karen Bruns° and Klaus G. Grunert (1995) ,"Development and Testing of a Cross-Culturally Valid Instrument: Food-Related Life Style", in NA - Advances in Consumer Research Volume 22, eds. Frank R. Kardes and Mita Sujan, Provo, UT : Association for Consumer Research, Pages: 475-480.

Advances in Consumer Research Volume 22, 1995      Pages 475-480

DEVELOPMENT AND TESTING OF A CROSS-CULTURALLY VALID INSTRUMENT: FOOD-RELATED LIFE STYLE

Karen Bruns°, The Aarhus School of Business

Klaus G. Grunert, The Aarhus School of Business

ABSTRACT -

Based on a cognitive perspective, we propose to make life style specific to certain areas of consumption. The specific area of consumption studied here is food, resulting in a concept of food-related life style. We have developed an instrument that can measure food-related life style in a cross-culturally valid way. To this end we have developed a pool of 202 items, collected data in three countries, and have constructed scales based on cross-culturally stable factor patterns. We have then applied the set of scales to a fourth country, in order to further test the cross-cultural validity of the instrument.

PROBLEMS WITH LIFE STYLE INSTRUMENTS IN MARKET SURVEILLANCE

In the market surveillance of consumers, life-style instruments have been used in order to detect major trends over time and/or in order to analyse differences and similarities across markets in the search of global or, eg, pan-European segments. Most life style studies, commercial and academic, follow a common pattern. They are based on a large battery of AIO items that are reduced analytically to few, usually only two, dimensions. The resulting space, sometimes called an attitude map or a value map, is then used to classify consumers on the remaining dimensions, which leads to life style segments. This type of life style research has been criticized on several grounds (eg, Anderson and Golden 1984; Askegaard 1993; Banning 1987; Lastovicka 1982; Roos 1986). Apart from the general terminological confusion about the term life style (Anderson and Golden 1984), it has been criticized that the instruments are not guided by theory. Life style types come about based on dimensions derived by exploratory data analysis techniques like factor analysis or correspondence analysis. These techniques are applied to sets of items, the generation of which is not theoretically guided either, but is very much based on common sense reasoning and implicit experience in carrying out market research. While such a research procedure may be appropriate in the early phase of the life cycle of a research technique, one should hope that, based on such exploratory analysis, theory should develop, which could then guide the analysis of new and better measurement instruments. Also, many feel that consumer behaviour is such a well-researched area that it should be possible to obtain some theoretical input from there that could enrich life style research. Another major criticism, and one we explicitly will take up in this paper, is that the cross-cultural validity of the international life style instruments remains to be demonstrated. The larger pan-European life style studies like RISC and CCA [RISC: Research Institute of Social Change; CCA: Centre de Communication Avance] provide data which aim at identifying similar life style segments across borders, and numerous other life style studies have tried to identify cultural differences in life style (eg, Douglas and Urban, 1977; Hui, Joy, Kim and Laroche 1990; Laroche et al. 1990; Linton and Broadbent 1975). Collecting data in different cultures with the aim of obtaining comparative results requires that the measurement intrument has cross-cultural validity, ie, that translation and measurement equivalence are ensured or at least tested (cp. Chandran and Wiley 1987; Green and White 1976; Sekaran 1983). For the kind of data involved in life style studies, factor invariance is a good criterion for investigating the degree of translation and measurement equivalence actually achieved. Various degrees of factor invariance can be distinguished, corresponding to various degrees of cultural comparability (S.C. Grunert, K.G. Grunert, and Kristensen 1992). However, such investigations have not yet been reported for life style data.

We have tried to develop a life style instrument that makes progress with these problems. We will briefly present a theoretical framework that can guide us in the formulation of candidates for life style items, show how we have selected items that are cross-culturally valid, based on an extended factor congruence criterion, and test the selected items in a new cultural setting for further cross-cultural validation.

A COGNITIVE APPROACH TO LIFE STYLE

We regard life style as a mental construct which explains, but is not identical with, actual behaviour, and define life style as the system of cognitive categories, scripts, and their associations, which relate a set of products to a set of values. This proposed definition warrants a number of comments.

(i) It makes life style distinct from values, since values are self-relevant and provide motivation, while life style links products to self-relevant consequences, ie, values.

(ii) Life styles transcend individual brands or products, but may be specific to a product class. Thus, it makes sense to talk about a food-related life style, or a housing-related life style.

(iii) Life styles are clearly placed in a hierarchy of constructs of different levels of abstraction, where life styles have an intermediate place between values and product/brand perceptions or attitudes.

STUDY I: DEVELOPING A CROSS-CULTURALLY VALID INSTRUMENT TO MEASURE FOOD-RELATED LIFE STYLE

We have argued that life style may be specific to a product class. The product class we have chosen for the current project is food products. Hence, we aim to develop a cross-culturally valid instrument to measure food-related life style (FRL). How are food products related to values in consumers' cognitive structure? It may be possible to distinguish relevant parts or aspects of cognitive structure, which may then be the starting point for item formulation. In figure 1, an attempt is made to delineate relevant parts of cognitive structure, and how they contribute to linking food products to values. The boxes indicate groups of cognitive categories, and the lines associations between them.

Shopping scripts. How do people shop for food products? Is their decision-making characterized by impulse buying, or by extensive deliberation? Do they read labels and other product information, or do they rely on the advice of experts, like friends or sales personnel? In which shops - one-stop shopping versus speciality food shops?

Meal preparation scripts. How are the products purchased transformed into meals? How much time is used for preparation? Is preparation characterized by efficiency, or by indulgence? Is it a social activity, or one characterized by family division of labour? To which extent is it planned or spontaneous?

FIGURE 1

A COGNITIVE STRUCTURE MODEL FOR FOOD-RELATED LIFE STYLE

Desired higher-order product attributes. This refers not to concrete attributes of individual products, but to attributes which may apply to food products in general. Examples may be healthy, nutritious, natural, convenient.

Usage situations. What are "the" meals? How are they spread over the day? Is a meal perceived differently when eaten alone, rather then with the family?

Desired consequences. What is expected from a meal, and what is the relative importance of these various consequences? How important is nutrition compared to the social event? How important is hedonism (cf. S.C. Grunert 1993).

We have then developed a survey instrument by applying the following procedure:

! Generation of a pool of items covering the five elements of food-related life styles based on the theoretical foundation.

! Collection of data, using the item pool, in three European countries.

! Exploratory factor analysis, within each of the five elements, and separately for each country.

! Search for factors which seem to be stable across the three samples.

! Construction of scales for each of the remaining factors. Analysis of scale reliability across and within samples. Modification of scales with the aim to retain three items per scale.

! Testing the set of cross-cultural factors by confirmatory factor analysis.

We have then applied the instrument obtained in a fourth country for further cross-cultural validation. This will be addressed in the next section.

Generation of item pool

202 Likert-type items were constructed which covered the five constructs of food-related life style defined above (Shopping scripts, desired higher-order product attributes, meal preparation scripts, usage situations, and desired consequences). Inspiration for formulating these items was drawn from the food choice literature, food journals, women's magazines, and earlier life style studies like RISC. Agreement with the items had to be rated on a five-point scale. The items were originally formulated in English and subsequently translated into Danish and French. Items were arranged by the five constructs in the questionnaire; the sequence in which the five constructs appeared was varied at random in the questionnaire.

Data collection

Data were collected in Denmark, England, and France. Given the nature of this study, sampling was done not with respect to representativeness, but with respect to obtaining three samples which would be as homogeneous as possible. The target population was defined as married women with children at school age living in metropolitan areas. In each country, one metropolitan area was selected, viz. Copenhagen, London, and Paris. In each of these areas schools were selected (2 in Copenhagen, 4 in London, 3 in Paris), with the aim of soliciting the co-operation of teachers in asking school children to take a questionnaire home to their mothers. In each country, 300 questionnaires were distributed in this way. The response rates were 78% in Denmark, 47% in England, and 32% in France, resulting in sample sizes of 233, 139, and 94.

Exploratory factor analysis and construction of scales

For each sample and for the five groups of items, separate exploratory factor analyses (principal component analysis, varimax rotation) were carried out. The aim of this procedure was to check whether items would tend to group together in similar factors across the three samples. Factor congruence across cultures is a major indicator of a cross-culturally similar interpretation of the items (S.C. Grunert et al. 1992). 21 factors were identified which seem to appear across the three samples. They were:

! shopping scripts: importance of product information, attitude towards advertising, joy of shopping, speciality shops, price criterion, shopping list

! higher-order product attributes: health, price-quality-relation, novelty, organic products

! meal preparation scripts: involvement with cooking, looking for new ways, convenience, whole family, spontaneity, women's task

! usage situations: snacks versus meals, social event

! desired consequences: self-fulfilment in food, security, social relationships.

Items with high loadings on these factors in at least two of the three samples were combined into scales. Scale reliabilities (Cronbach's alpha) were computed, and where scales contained more than three items, the items which gave the highest reliability for the pooled data were retained. Subsequently, the reliabilities were checked also for the three samples separately. Results from these analysis can be found in Grunert, Bruns° and Bisp (forthcoming). Not all the resulting scales were satisfactory. In a number of cases, only two suitable items were found for a scale. This has lead to some modifications in the study described in the next section. [The complete set of scales can be obtained from Karen Brunso -fax +45 86 15 39 88 or e-mail kab@hdc.hha.dk.]

Confirmatory factor analysis

Confirmatory factor analysis can be used to find out whether a set of data is compatible with a pre-specified factor structure. It can also be applied to multiple samples and can then be used to check whether the data are compatible with the assumption that the factor structure in the samples is the same. Factor invariance has often been suggested as a validation instrument in cross-cultural research. S.C. Grunert et al. (1992) have recently suggested that several levels of factor congruence may be distinguished and have related these levels to different degrees of cultural compatibility based on a cognitive view of cultural differences. The basic argument is as follows: If we have a vector of measures which, like in the present study, is taken as indicators of a smaller set of underlying latent variables, then we have, in LISREL notation,

    X=Lx+d

   and

    S=LFL'+Qd

with X a vector of measured values, L a matrix of factor loadings, x a vector of factor scores, d a vector of error terms, S the covariance matrix of the measured values, F the covariance matrix of the factor scores, and Qd the covariance matrix of the error terms. When discussing factor congruence, the common interpretation is that the matrix of loadings in two samples has the same pattern, ie, the same non-zero elements. However, this is obviously only the weakest form of comparability between two sets of data. A stronger form of comparability would exist when the matrix of loadings were in fact identical, since this seems to indicate that the way in which the measurement items relate to underlying constructs was in fact the same across samples. This would still allow, however, differences in how the factors are correlated in the two samples, and differences in error, ie, in the reliabilities of the individual items. Identical correlations between the factors would strengthen our confidence in that the factors do in fact tap the same sets of meanings in different cultures, whereas identical item reliabilities would strengthen our belief in that the individual items in fact were perceived (cognitively processed) in the same way. In cross-cultural research, which usually involves translation and therefore mapping questionnaire items from one set of cognitive categories into another, we would not usually expect item reliabilities to be the same. Thus, four levels of cross-cultural comparability can be distinguished (S.C. Grunert et al. 1992):

    ! Li and Lj have the same pattern: minimal cultural comparability

    ! Li=Lj: weak cultural comparability

    ! Li=Lj, and Fi=Fj: strong cultural comparability

    ! Li=Lj, Fi=Fj, and Qdi=Qdj: weak cutural identity

The strongest condition is called weak cultural identity, because the only way in which the samples can differ is in the level of endorsement of the various items, while everything else - their complete meaning structure, including item reliabilities - is the same. When also the levels of endorsement are the same, one would talk about strong cultural identity. For each of the five areas of food-related life style, the items retained were entered into confirmatory factor analyses corresponding to the four levels of cultural comparability described above. The results can be seen in table 1. Several measures of fit are given: for each sample, the goodness of fit index GFI and the root mean square residuals RMR, and for the set of three samples the X2 value and the degrees of freedom. Since the X2 value is vulnerable to sample size, a rule of thumb is to look at the size of X2 relative to the degrees of freedom eg, by dividing X2 with degrees of freedom. The GFI measure indicates the relative amount of variances and covariances accounted for by the model, a value of .9 or more indicates that the model fits the data well (Bagozzi and Yi 1988). Furthermore, the GFI measure is independent of sample size, and can be used for comparison of different models on the same data and on different data. The third measure, RMR, is a measure of the average of the fitted residuals, and can be used for comparison of different models on the same data. RMR should below (Bagozzi and Yi 1988).

The results from the analysis can also be used to detect weak items. This can be done by inspecting the estimated loadings, the item reliabilities, and the modifications indices for loadings forced to be zero. The following results emerged:

Shopping scripts. The tests yield results which indicate that the criteria of minimum and weak cultural comparability can be accepted at least for the Copenhagen and Paris samples. The values for the London sample are worse, and a number of problematic items were identified.

Higher-order product attributes. All four tests show a reasonably good fit, but weak cultural comparability seems again to be the best description across the three samples, considering especially the behaviour of the RMR in the Paris sample.

Meal preparation scripts. The fit is not as good as for the other aspects of FRL, and the fit for the Danish sample is notably better than for the other two. A main problem seems to reside in the convenience scale, and some items were correspondingly changed for the next study.

Usage situations. The data have generally a high degree of comparability. All datasets fulfill the conditions of weak cultural comparability, and even for the two more stringent criteria the fit indices are still rather acceptable.

Desired consequences. As for usage situations, the data have generally a high degree of comparability. All datasets fulfill the conditions of weak cultural comparability, and even for the two more stringent criteria the fit indices are still rather acceptable.

TABLE 1

CONFIMATORY FACTOR ANALYSIS, STUDY 1

In general, the results of the confirmatory factor analysis show that the scales developed are a promising starting point for the development of a cross-culturally valid instrument to measure food-related life styles. For all five elements of food-related life style, at least the level of weak cultural comparability was obtained. However, the analysis also pointed at certain scale items which had to be improved.

STUDY II: CROSS-CULTURAL VALIDATION IN A NEW CULTURAL SETTING

The first study allowed us to identify 21 cross-culturally valid scales. We then improved the items based on the results of the previous analysis and filled up with new items so that all 21 scales had 3 items. In addition, discussions with users of the instrument led us to include two additional scales for abstract product attributes, namely freshness and tastiness. We then went into the first full-scale application of the instrument, ie its application with a representative random sample. We chose to carry out this application in a fourth cultural setting, Germany, in order to find out whether the data are compatible with the structure emanating from study I.

Sample and data collection

The population was defined as persons of age 16 and above who have the main responsibility for buying food and preparing meals in the household. A representative sample of 1000 respondents was obtained by a random-route procedure. Data collection was carried out by personal interviews.

Confirmatory factor analysis

In order to test to what degree the structure in the German data is comparable with the structure in the data from the first study, confirmatory factor analysis was conducted with the new data. For minimum cultural comparability, this was done by imposing the pattern of factor loadings corresponding to the factor pattern resulting from study I. For the higher levels of cultural comparability, this was done by fixing the respective parameters (factor loadings, factor correlations, and item reliabilities) to the values estimated in study I for these parameters at the respective level of comparability.

The results can be seen in table 2. It can be seen that the RMR values are higher compared to the first study, which may be explained by the fact that the measurement scale used in the second study was extended from a 5 to a 7-point scale. The size of the residuals vary with the scale of the variables; changing the unit of measurement of a variable will change the variances and covariances, and thus the size of the residuals. For this reason, RMR will only be evaluated if it is considerably higher than average for a construct given the level of cultural comparability, or if it makes a jump in value (cf. Bagozzi and Yi 1988).

Except for the construct usage situations it has not been possible to impose a complete data structure on the new data, because new items were developed for the German sample. For the new items, only the loading pattern could be specified. The following results can be obtained from the cross-cultural comparability analysis:

Shopping scripts. The model fit is acceptable for both minimum and weak cultural comparability, after which there is a drop for all three measures of fit. One item has been added to this construct compared to study I; analysis of model fit without this item does not change the result.

Higher-order product attributes. Again both minimum and weak cultural comparability can be accepted. From this level, the model fit drops dramatically, and RMR can not be computed due to failed admissibility test. For this construct 10 new items were developed, and deletion of some of these can improve model fit.

TABLE 2

CONFIRMATORY FACTOR ANALYSIS, STUDY II

Meal preparation scripts. Here the values of RMR are somewhat higher than for the two previous constructs, while the two other measures are about the same. Based on the measures of X2 and GFI, this construct also fulfill the criteria for minimum and weak cultural comparability.

Usage situations. This construct has the best model fit at the minimum and weak cultural comparability levels with respect to GFI, and also with respect to X2 at the weak cultural comparability level. This is, as mentioned, the only construct which is completely identical to the first study.

Desired consequences. The results suggest that only minimum cultural comparability can be accepted here, due to a considerable drop/jump in both X2 and RMR after this level.

When comparing the results from tables 1 and 2, it has to be kept in mind that table 2 represents a much stronger test. In table 1, the data were tested against a structure developed on the same data, and all parameters, even when forced to be equal across samples, were estimated for optimal fit. In table 2, the same structure is applied to new data, and the fit is estimated based on parameter estimates from the first study.

Generally, none of the factors can be said to be culturally identical or to be strongly culturally comparable. On the other hand, most of the constructs can be said to meet the criteria for minimum and weak cultural comparability. This result supports that the items have been interpreted in a similar way in Germany compared to the first study, and yield support for the instrument as being cross-cultural valid. Further validation should take place though, with adjustments of the instrument, in a new cultural setting.

CONCLUSIONS

Starting from the need to develop an efficient instrument for the surveillance of consumers on export markets, we have attempted to present a new view of life style, based on a cognitive perspective, which makes life style specific to certain areas of consumption. The specific area of consumption studied here is food, resulting in a concept of food-related life style. We have tried to develop an instrument that can measure food-related life style in a cross-culturally valid way. To this end, we have collected a pool of 202 items, collected data in three countries, and have constructed scales based on cross-culturally stable factor patterns. These scales have then been subjected to a number of tests of reliability and validity.

We have then applied the set of scales to a fourth country, Germany, based on a representative sample of 1000 respondents.

When imposing the data structure obtained from the first study on the new data by confirmatory factor analysis, it was found that at least minimum cultural comparability exists for all scales.

The present research would benefit from more replications in additional countries, leading to more evidence for the cross-cultural validity of the scales developed. In addition, research relating these scales to behavioural variables, thus providing some evidence for its applicability in terms of prognostic validity, is clearly called for.

REFERENCES

Anderson, Thomas W. and Linda L. Golden (1984),"Life style and psychographics: A critical review and recommendation," in Advances in Consumer Research, ed. T.C. Kinnear Vol. 11, pp. 405-411. Provo, UT: Association for Consumer Research.

Askegaard, S°ren (1993),"Livsstilsbegrebet: Problemer og muligheder," Ledelse og Erhvervs°konomi, 57, 91-102.

Bagozzi, Richard and Youjae Yi (1988),"On the Evaluation of Structural Equation Models," Journal of the Academy of Marketing Science, 16 (1), 74-94.

Banning, T.E. (1987), Lebensstilorientierte Marketing Theorie, Heidelberg: Physica.

Chandran, Rajan and James B. Wiley (1987),"Instrument equivalence in cross-cultural research," in Proceedings of the Second Symposium on Cross-cultural Consumer and Business Studies, eds. C.F. Keown & A.G. Woodside, pp. 108-112. Honolulu, HA: University of Hawaii.

Douglas, Susan P. and Christine D. Urban (1977),"Life-style analysis to profile women in international markets," Journal of Marketing, 41(3), 46-54.

Green, R. D. and White, P. (1976),"Methodological considerations in cross-national consumer research," Journal of International Business Studies, 7, 81-87.

Grunert, Klaus G., Karen Bruns° and S°ren Bisp (forthcomming),"Food-related life style: Development of a cross-culturally valid instrument for market surveillance," in: L. Kahle & C. Chiagouris (Eds.), Values, lifestyles, and psychographics. Hillsdale, NJ: Erlbaum.

Grunert, Suzanne C. (1993), Essen und Emotionen. Weinheim, PVU.

Grunert, Suzanne C., Klaus G. Grunert and Kai Kristensen (1992),"The cross-cultural validity of the List of Values LOV: A comparison of nine samples from five countries," in Developments and applications in structural equation modelling, eds. J.J.G. Schmeets, M.E.P. Odekerken & F.J.R. van de Pol, pp. 89-99. Amsterdam: Sociometric Research Foundation.

Hui, Michael, Annamma Joy, Chankon Kim and Michel Laroche (1990),"Differences in lifestyle among four major subscultures in a bi-cultural environment," in Proceedings of the Third Symposium on Cross-cultural Consumer and Business Studies, eds. C. F. Keown et al., pp. 139-150. Honolulu, HA: University of Hawaii.

Laroche, Michel, Ron McTavish, Lester Johnson, Annamma Joy, Chankon Kim and Susan Rankine (1990),"Consumption patterns and lifestyle differences betweeen English-French Canadian and Australian consumers," in Procedings of the Third Symposium on Cross-cultural Consumer and Business Studies, eds. C. F. Keown et al., pp 151-160. Honolulu, HA: University of Hawaii.

Lastovicka, John L. (1982),"On the validation of life style traits," Journal of Marketing Research, 19, 126-138.

Linton, Anna and Simon Broadbent (1975),"International life style comparisons: An aid to marketers,"Advertising Quarterly (Summer), 15-18.

Roos, J.P. (1986),"On way of life typologies," in Environmental impact of consumption patterns, ed. Liisa Uusitalo, pp. 38-55. Aldershot: Gower.

Sekaran, Uma (1983),"Methodological and theoretical issues and advancements in cross-cultural research," Journal of International Business Studies, 14(3), 61-73.

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