5382089

You might also like

You are on page 1of 16
Retail services image measurement: an examination of the stability of a numerical comparative scale R. Kenneth Teas Abstract ‘The numerical comparative scale (NCS) is a multi-attribute measurement procedure in which the respondent provides simultaneous comparative ratings of multiple objects (ie., alternative retail stores, products, etc.) using bipolar scales similar to semantic differential scales. Recent research indicates. that scales requiring respondents to rate multiple retail stores on an attribute-by- attribute basis may be vulnerable to rating instability problems resulting from rating context effects. The purpose of this paper, therefore, is to examine the evidence of this potential rating instability problem and to report the results, of an experiment designed to test the numerical comparative rating scale under conditions that potentially produce context-induced rating instability ‘The findings indicate the scale is subject to problems of instability resulting from the multiple-store rating context. Based upon the findings, potential methods of increasing the stability of the scale are explored. Keywords Retail, consumers, semantic differential, image measurement. Introduction ‘The extensive use of the semantic differential scale in multi-attribute image measurement has resulted in researchers proposing a number of modifications of the scale. Three modified scales that have been pro- posed for retail store image measurement, which are described in Figures 1 and 2, are the graphic positioning scale (Narayana 1977; Swan and Futrell 1980; Stem and Noazin 1985), the upgraded semantic dif- ferential scale (Evans 1980), and the numerical comparative scale © Routledge 1994 Copyright © 2001. All Rights Reserved. 428 R. Kenneth Teas (Golden, Albaum and Zimmer 1987). The primary reason for develop- ing these scales is to enable the efficient scaling of more than one retail store by cach respondent. The primary advantages proponents of these kinds of scales have identified are: 1. Less space is required on the questionnaire than is required by stan- dard semantic differentials. Therefore, the length of mail survey questionnaires can be reduced (Golden et a/. 1987). 2. The rating of competing retail stores comparatively on an attribute by-attribute basis can be expected to result in increased rating dis- crimination (Swan and Futrell 1980; Narayana 1977). 3. By shifting the respondent’s focus from (a) one object’s ratings on a set of image questions to (b) multiple objects’ comparative ratings on an attribute-by-attribute basis, response halo effects can be reduced (Wu and Petroshius 1987), Halo effects involve the tendency of a respondent ‘to rate individual traits according to the rater’s general impression of the object (e.g., a store) being rated’ (Wu and Petroshius 1987), That is, a respondent's ratings of individual traits are ‘colored? by a general preference or ‘affective overtone’ (Holbrook 1983) Although tests of these modified semantic differential scales indicate the scales may be valid and reliable (Golden et al. 1987; Swan and Futrell 1980), other tests of scales that require respondents to rate mul- tiple objects on an attribute-by-attribute basis indicate the ratings of a particular object are affected by the other objects being rated. For example, research results reported by Teas and Wong (1992a, 192b) indicate that, when respondents rate multiple retail stores, the rating of a particular store (the target store) on an attribute, such as merchandise quality, can vary depending upon the other stores (context stores) that are being rated. Furthermore, the results indicate that the standard methods used to control for these ‘context store’ effects, such as ran- domizing the order in which the stores appear on the questionnaire, may not always be effective, Consequently, the purpose of this study is to examine the degree to which one modified semantic differential scale ~ the numerical comparative scale (Golden ef a/, 1987) ~ is vulnerable to multiple-store rating context effects. In this study muhiple-store rating context effects involve the degree to which the ratings of a particular store (the target store) are affected by the context store with which the target store is rated. Multi-entity scaling via the numerical comparative rating scale Multi-entity rating scales Multi-entity rating scales are scales on which two or more entities are rated by individual respondents. Two general types of multi-entity rat- Copyright © 2001. All Rights Reserved. Retail services image measurement 429 1. Graphic Positioning Scale - Each respondent rates a set of stores on a series of scales using store codes (e.9., S = Sears, W = Wal-Mart, P = J.C. Penney) to indicate the placement of stores on a graphic scale: P Convenient WT] | || | | HHI It | | Inconvenient Location | Mi Location Low | nati mm High Pres | HI Pres Pow Fast I i| Slow ee HTHVIAUTIRVLITIIIVTIVIU Soe 2. Upgraded Semantic Differential - Each respondent rates a set of stores on a series of scales by using store codes to indicate the placement of stores on a numerical scale: Convenient Wo PS Inconvenient Location 7 6 5 4 3 2 1 Location Low wopios High Prices 765 45 2 4. Prices Fast 8 Pow Slow Service 765 4 3 2 4. Service Figure 1 Modified semantic differential scale Copyright © 2001. All Rights Reserved. 430 R. Kenneth Teas (1.1) The Numerical Comparative Scale DIRECTIONS: This section of the questionnaire concerns your opinions about three retail stores in (city) - (store X), (store Y), and (store Z). Please indicate your opinions of each of the stores using the scale below. Write the number corresponding to your opinion in the blank under each store name. For example, suppose you were to describe each store on the following scale: be Loree : Store X) (Store Y) Gur) Dark 1 1204567 Bins eT 2. ee ee A’‘4’ for Store X means you believe (Store X) is neither bright nor dark; a7’ for (Store Y) means you think (Store Y) is extremely bright; and a ‘2'for Store Z means you think (Store 2) is quite dark. For each scale below write in the blank the number that best describes your impression of (Store X), (Store Y), and (Store Z): _» Convenisnt (Store X aoe Figure 2 Multi-entity bipolar rating procedures ing scales are attribute-based and entity-based scales which are defined by Teas and Wong (1992a) as follows: Attribute-based Multi-entity Scaling is a multiattribute measurement procedure in which cach of two or more entities are rated simultane- ously by the respondent on cach attribute. That is, all the entities are Copyright © 2001. All Rights Reserved. Retail services image measurement 431 rated on the first attribute together, followed by ratings on the second attribute, and so on. Entity-based Multientity Scaling is a multiattribute measurement pro- cedure in which each of two or more entities are rated separately on the set of attributes. That is, the respondent rates one of the entities on all the attributes prior to rating the second entity on all the attrib- utes, and so on. Each of the modified semantic differential scales described in Figures 1 and 2 requires respondents to rate multiple retail stores simultane- ously on each attribute. Consequently, the scales can be classified as attribute-based multi-entity rating scales. In contrast, when traditional semantic differential scales are used by a respondent to rate multiple stores, the scaling procedures involve entity-based scaling ~ each store is rated separately on a set of semantic differential scales. Golden er al. (1987) proposed and tested a numerical comparative scale (NCS) which is an attribute-based multi-entity rating procedure in which two or more retail stores are rated together on a set of numeri- cally labelled bipolar scales. The general format of the scale is presented in Figure 2. As indicated in the figure, the NCS produces considerable questionnaire-length economy because the multiple stores are rated together on a single sct of scale items. In addition, empirical tests reported by Golden et al, (1987) indicate that the NCS, when compared with the graphic positioning scale (Narayana 1977), resulted in higher questionnaire-completion rates, lower measurement error and lower item non-response rates. An issue that has not been addressed with respect to comparative semantic differential scales, such as the NCS and the graphic positioning scale, is the degree to which they are vulnerable to multiple-store rating context effects. The context effect problem Alternative multi-entity scaling procedures are often assumed to pro- duce similar information (Narayana 197); consequently, researchers often base scale-selection decisions on data-collection efficiency criteria such as: the amount of information that can be obtained from an indi- vidual respondent; the speed of response; and the required questionnaire length (Swan and Futrell 1980). However, the results of research com- paring multi-attribute scaling methods (attribute-based, entity-based and single-entity scaling) indicate that different scaling methods produce dif- ferent results (Jaffe and Nebenzah! 1984; Teas and Wong 1992a) and that attribute-based scaling is particularly vulnerable to multiple-entity rating context effects — the rating of a particular entity on an attribute tends to vary depending upon the other entities being rated Copyright © 2001. All Rights Reserved. 432. R. Kenneth Teas Several theoretical explanations have been suggested as explanations for these context effects, such as perceptual contrast and scale-anchoring effects. Perceptual contrast effects involve a process in which the respon- dent’s perception of a target object (i.c., a particular retail store) is affected by the context (e.g., other retail stores) with which the target object is compared. For example, when a particular target retail store is evaluated in the context of particularly excellent stores, perceptual con- trast effects would cause the respondent's evaluation of the target store to shift in the negative (non-excellence) direction. In general, these effects result in the respondent's evaluation of the target store being dis- placed away from his/her evaluation of the context stores. Response scale anchoring involves a process in which respondents align the rating scales according to the evaluated range of the set of objects being rated (Sherman et al. 1978). Response scale anchoring does not involve modi- fied judgements, but rather modifications of the way the rating scale is used to reflect judgements, Two potential methods of controlling for perceptual contrast and scale-anchoring effects involve using randomiza~ tion procedures in questionnaire construction and analysing variable relationships rather than absolute scores. Randomization The order of presentation of the individual question naire measurement items is often randomized to avoid order bias. In retail image studies, therefore, the order in which the stores are rated could be randomly varied across respondents, The primary rationale underlying this procedure is that, although individual responses may shift because of presentation order, randomizing the order across ques- tionnaircs will result in shifts cancelling out and, therefore, the mean responses will be unbiased. This randomization procedure would be expected to be most effective if each respondent rates the entire set of stores being studied because the content of the rating context would be held constant across respondents (Teas and Wong 1992a).' However, this is often not possible. For example, the set of stores to be rated is often large; consequently, it may not be feasible for individual respon- dents to rate the entire set because of questionnaire/ interview length limitations, Furthermore, individual respondents may not be knowledge- able about all of the stores being rated. Consequently, they are asked to rate only those stores with which they are familiar. Because of the ques tionnaire length and a respondent's familiarity limitations, respondents often rate only a subset of stores with different respondent's rating dif- ferent subsets. Consequently, differences among respondents may be due to context effects rather than being true differences in attitudes and opinions. Variable relationships Schuman and Presser (1981) note that, although the absolute values of responses to questionnaire items (i.e. “item mar- ginals’) are affected by context effects, researchers often assume the relationship among variables will not be affected by context effects. For Copyright © 2001. All Rights Reserved. Retail services image measurement 433 example, when using NCS rating procedures the assumption would be that, although a store’s ratings on an image scale may shift because of context effects, the differences between two customer groups’ ratings (cg., customers versus non-customers) would not be affected. However, research results reported by Schuman and Presser (1981) suggest that the assumption that context may affect absolute scores but not estimated comparative rating differences may be incorrect because response shifts due to context effects may not be uniform across respondents and, therefore, across groups. Consequently, estimates of the magnitude of group differences can be biased by context effects. In summary, because of the empirical evidence that attribute-based rating scales are vulnerable to context-induced rating instability, and the difficulties associated with traditional remedies for the problem, it is important to test the stability of alternative multi-entity rating scales. ‘The purpose of this study, therefore, is to examine the degree to which the numerical comparative scale (NCS) is vulnerable to context-induced rating instability. Based upon the published research evidence concern- ing the stability of attribute-based scaling, the following hypotheses are examined: HI: Numerical comparative sealing of a target retail store is affected by the context store with which the target store is rated. he following two methods of obtaining a target store’s multi- attribute image rating will produce different results ~ (a) the tar- get store’s ratings obtained along with a context store’s ratings using the numerical comparative scaling procedures, and (b) the target store’s ratings obtained without a context store’s ratings using the numerical comparative scaling procedures H2: The method Experimental design An experiment involving the measurement of retail store service image was designed to examine the research questions. Preliminary exploratory research was used to identify three stores that were differentiated in terms of university students’ perceptions of overall service and mer- chandise quality. Based upon this research, the stores were ranked in terms of overall quality. ‘The first and third ranked stores (stores A and B, respectively) were selected to be the experimental treatment context stores and the second ranked store was selected to be the target store. ‘The experiment involved collecting survey data using numerical com- Parative scaling (NCS) procedures according to the three-cell experimen- tal design presented in Table 1. Numerical comparative scaling was used in cells 1 and 2 with the store A included as the context store for cell 1 Copyright © 2001. All Rights Reserved. 434 R. Kenneth Teas ‘Table 1 Experimental design Cell ” Measures Context store Target store 1 62 A 2 65 B 3 67 Nowe The five sets of measures are: Gell 1; Context store (A) and target store (Ty) measures are ‘obtained via the Numerical Comparative Rating scale. Cell 2: Context store (B) and target store (Th) measures. are obtained via the Numerical Comparative Rating scale Cell 3: Target store (T,) control measures are obtained in the absence of any. context store using the Numerical ‘Comparative Rating scale. respondents and store B included as the context store for cell 2 respon dents. In cell 3, the control cell, the numerical comparative scale was used to obtain target store ratings without any context store. In cells 1 and 2, the target store is the second store listed on all questionnaires. Measurement ‘The NCS questionnaire format is presented in Figure 2. The specific items used in the questionnaire were derived from the literature on retail image measurement (Pessemier 1980; Pathak, Crissy and Sweitzer 1974-5; Marks 1976; Golden et a/. 1987; Hirschman, Greenberg and Robertson 1978; Nevin and Houston 1980; Berry 1969; Gautschi 1981; Jolson and Spath 1973; McDougall and Fry 1974-5; Swan and Futrell 1980; Wu and Petroshius 1987; Menezes and Elbert 1979; Zelnio and Gagon 1981; Hauser and Koppelman 1979; Teas and Wong 1992a). A total of thirty-three items was used. Issues measured included store per- sonnel, physical facilities, retail service, operating hours, merchandise, price, promotions, parking, store layout, checkout lines, location conve- nience, store hours and merchandise display. The sample “The experimental subjects consisted of 194 undergraduate business stu- dents from a midwestern university. These students were randomly assigned to the thrce cells of the experiment which resulted in 62, 65 and 67 student responses for cells 1, 2 and 3, respectively. The students filled out the questionnaires during class meetings in groups of 20 to 44 students.” Copyright © 2001. All Rights Reserved. Retail services image measurement 435 Hypotheses test results Hypothesis 1 suggests that changing the context store that is being rated with the target store will result in different multi-attribute target store ratings. Consequently, the hypothesis was tested using multivariate analysis of variance (MANOVA) to compare the multi-attribute ratings of the target store provided by the respondents in the two experimental cells. The MANOVA results, which are presented in Table 2, indicate statistically significantly (p < .004) different target store ratings in cells 1 and 2. This finding supports hypothesis H1. Hypothesis 2 suggests that the multi-attribute ratings of the target store obtained in cells 1 and 2 will not be equal to the multi-attribute ratings obtained in the contro! cell 3. Consequently, the hypothesis was tested using MANOVA to test the statistical significance of the differ- ences between the cell | and cell 3 ratings and the differences between the cell 2 and cell 3 ratings. The results, which are presented in Table 2, indicate the target store ratings from the experimental cells 1 and 2 are statistically significantly different from the target store ratings from the control cell 3. These findings support hypothesis H2. ‘Table 2 Multivariate analysis of variance results (2a) Test of ‘hypothesis Hl: MANOVA test of the difference between the target store ratings from each of the two experimental groups (Cells 1 and 2) Celis compared Fr Degrees of freedom Significance of F 1 and 2 2.022 33/93 004 (2b) Tests of hypothesis H2: MANOVA tests of the differences between (a) the target store ratings from each of the two experimental groups and (b) the target store ratings from the controt group Cells compared F Degrees of freedom Significance of F Vand 3 2.235 33/98, 001 2and3 1.943 33/95 007 Follow-up analysis The statistical significance of the MANOVA tests indicates the context store treatments in experimental cells 1 and 2 affected the multi- attribute target store ratings. The purpose of the follow-up analysis is to examine the nature of these effects. Since multi-entity ratings are often used to produce competitive positioning maps, multiple discriminant analysis maps’ were used in the follow-up analysis Five sets of store ratings were used in the multiple discriminant analysis ~ two sets of ratings from each experimental cell and one set of Copyright © 2001. All Rights Reserved. 436 R. Kenneth Teas ratings from the control cell, Consequently, the multiple discriminant map provides store-image positions corresponding to three sets of the target-store ratings (from cells 1, 2 and 3) and one set of ratings for cach of the two context stores. The control-cell target-store ratings are used to provide a context-free benchmark rating for the target store “The results of the multiple discriminant analysis, which are presented in Table 3, indicate four statistically discriminant functions (p < .05).* ‘The estimated structure coefficients (e.g., loading coefficients), which were rotated via varimax rotation, are also presented in Table 3. As indicated in the table, the variables with the highest loading coefficients on the first function involve merchandise quality and the degree to which the store is prestigious, the physical facilities are visually appeal- ing and the prices are high. Based upon the content of these items, the first discriminant function is labelled ‘upscale’. The variables with the highest discriminant loading coefficients on the second discriminant function involve product selection and sales and promotion. This func- tion is labelled ‘merchandise’. The loadings on the third i function involve checkout issues and the degree to which the customer receives personal attention and prompt service. This function, therefore, jis labelled ‘transaction effectiveness’, ‘The variables loading on the fourth function involve helpful customer service provided by store employees. This function is labelled ‘employee helpfulness’. ‘Table 3 Multiple discriminant analysis results ~ percentage of variance explained by the four discriminant functions and the varimax rorated discriminant structure coefficients! Function 1(9 of variance = 54.41%) ~ “Upscale” 13. High (low) quality merchandise (.62) ‘A25. Prestigious (not a prestigious) store (.56) ‘AOL Physical facilities are (are not) visually appealing (.55) A24 High (low) prices (.53) Function 2 (% of variance = 27.17%) ~ ‘Merchandise’ AL4 Excellent (poor) product selection (.58) A16 Has (does not have) attractive sales and promotions (.38) Punction 3 (96 of variance ~ 9,90%) ~ “Transaction effectiveness 'A32._ Has (does not have) enough checkout clerks to handle customers during peak hours (.46) 27 Has short (long) checkout lines (.37) ‘409 Employees give (do nor give) you personal attention (.27) ‘804 You receive (do not receive) prompt service (24) Function 4 (% of variance = 8.52%) ~ 'Responsivencss’ ‘A38 Easy (difficult) co get questions answered (+4) ‘A03. When you have problems, the employees are (are not) sympathetic (41) ‘80S. Employees are always (never) willing to help you (39) Note 2 The percent of sariable explained and the structure coflcients are indicated in parentheses Copyright © 2001. All Rights Reserved. Retail services image measurement 437 The discriminant store centroids* are displayed in Figure 3. As indi- cated in the exhibit, some store rating instability is suggested by the results. First, on axis 1, the store A and store B centroid positions are located to the extreme right and extreme left portions of the discrimi- nant axis, respectively. The three target store centroids are located to the right of the store B location. The most dramatic target store posi- tion difference on the axis is the difference between the target-store rat- ings from the two experimental cells ~ (T,) and (Ty). The target-store ratings from cell 1, which produced centroid position (Ty), are based BT, Tp A - - Upscale To Tb Ta A 8 Merchandise BT, ite Ta OA Transaction 8 T A TT Effectiveness eee a Employee Helpfulness Note The discriminant function centroid values are used to position the stores on the four discriminant function axes. The store labels for the centroids are A = Context-store A centroid based upon data from experimental cell 1 B= Context-store B centroid based upon data from experimental cell 2 1, = Target-store centroid based upon data from experimental cell 1 — Store A is the context store. » = Target-store centroid based upon data from experimental cell 2 ~ Store B is the context store [= Target-store centroid based upon data from control cell 3. Figure 3 Mapping results: multiple discriminant centroid plots' Copyright © 2001. All Rights Reserved. 438 -R. Kenneth Teas upon target-store ratings that were obtained with context store B using NCS procedures. As indicated in Figure 3, the target-store centroid positions suggest possible contrast effects. That is, the (T4) centroid is shifted away from the store A centroid while the (Ty) centroid is shifted away from the store B centroid position. On the second discriminant axis, the store B position is located to the right of the axis while the store A position is located between the store B and the target-store positions, In contrast to the target-store centroid- position differences on the first discriminant function, the differences on the second discriminant function are small and indicate that the experi- mental cell target-store positions (I) and (Ty) are slightly more favourable than the control cell target-store positions. That is, the tar- get-store ratings from the experimental cells are slightly assimilated towards the positions of the context stores. On axis 3, the context-store A centroid position is located to the right of the axis and the context-store B centroid position is located to the left of the axis. The three target-store centroids are located between the two context-store positions. In contrast to the results on axis 1, the two experimental cell target-store positions reflect assimilation effects. For example, the target store centroid from cell 1 (T,) is shifted towards the context-store A centroid while the target-store centroid from cell 2 (Ty) is shifted towards the context-store B centroid. Furthermore, these posi- tion differences resulted in substantial differences among the three tar- get-store centroids. On the fourth discriminant axis, the store B centroid is located to the left of the axis. However, the store A centroid is located between the control cell target-store centroid (TF;) and the two experimental cell tar- get-store centroid locations (T, and ‘Ty). These differences in the target- store centroids indicate the experimental cell ratings resulted in substantial differences in the comparative locations with respect to the three target stores. First, the control cell results suggest that the target- store position (T.) is (a) inferior to the store A centroid position, and (b) very similar to the store B centroid position. However, the results from the two experimental cells suggest the target-store ratings (Tt and T,) are superior to both the context store B and context store A posi- tions. Discussion and conclusions ‘The numerical comparative scale (NCS) is an attribute-based_multi- entity rating scale in which each of two or more entities are rated simul taneously by the respondent on each attribute. Although a number of potential advantages of the scale (e.g., sampling efficiency, reduced halo effects, reduced questionnaire size and enhanced perceptual discrimina- Copyright © 2001. All Rights Reserved. Retail services image measurement 439 tion) are used to justify the use of attribute-based rating scales such as the NCS, the results of recent research indicate attribute-based rating scales may be vulnerable to context-induced rating instability. Consequently, the purpose of this study was to test, via a controlled experiment, the stability of a target store's NCS ratings when the target store is rated along with different context stores. The focus of the analysis concerned the degree to which different context stores result in different target store multi-attribute ratings. The experimental findings The results of multivariate analysis of variance (MANOVA) tests indi- cate the experimental manipulation of the context store with which the target store was rated via NCS procedures resulted in statistically signif- icant target-store rating differences. In addition, MANOVA tests pro- vided support for the hypothesis that the numerical comparative scaling of a target store that is rated with a context store during the rating process will not be equal to the numerical ratings of the target store that is rated without a context store during the rating process. Follow-up multiple discriminant analyses indicate the statistically sig- nificant differences in the target-store ratings across the experimental cells result in a considerable number of position differences with respect to the target entity’s multiple discriminant centroid locations on four multiple discriminant axes. Furthermore, these position differences appear to be the result of both contrast and assimilation effects. That is, some of the shifts involve the target store’s position in the perceptual space being shifted away from the context stores” positions (i.e., contrast effects). Other shifts involve the target store’s position in the perceptual space being shifted towards the context stores’ positions (ie., assimila~ tion effects). Some of the target-store centroid position shifts are quite dramatic and lead to conflicting interpretations. First, on the third multiple dis- criminant axis, the results of the positioning of the target store based upon the control group suggest the target store is positioned approxi- mately half way between the two context stores on the axis. On the other hand, when the target store is rated with store B using NCS pro- cedures, the target store appears to be almost undifferentiated from store B on axis 3. Similarly, when the target-store rating is obtained via the NCS in which store A is the context store, the positioning of the target store is quite close to store A on axis 3. Second, on axis 4, the target centroid differences result in different ordinal positions of the tar- get store centroids when compared to the context-store centroid posi- tions. That is, the data from the control cell 3 suggest the target-store position on the axis is between the centroid positions of stores B and A Copyright © 2001. All Rights Reserved. 440. Kenneth Teas with the target store being closer to the store B position. However, the data from the experimental groups result in the target-store centroids being positioned more favourably than stores A and B on the axis Practical implications Although this study was designed to magnify the potential instability problem using experimental manipulation of store rating context, the findings are suggestive of implications that are of considerable practical significance. If the primary marketing research question involves a need to collect context-free ratings of individual stores, the best data collec tion approach may be single-entity scaling where individual respondents rate only one entity. The classic way of conducting a study like this would be to survey equivalent random samples of respondents selected from the universe, one random sample for cach store being rated. Of course, this type of data collection is likely to be more costly than data collected via the numerical comparative scale because individual respon- dents are not rating multiple stores. Consequently, a larger number of respondents may be required to obtain a sufficient number of observa- tions for each store. The instability associated with the numerical comparative scale exam- ined in this study does not necessarily lead to the conclusion that researchers should always avoid collecting multiple store ratings from individual respondents. For example, if each respondent is able to rate the entire set of stores being compared, randomizing the order in which the stores are rated on the NCS may effectively control context-effect biases. However, when it is not possible for respondents to rate the entire set of retail stores, the findings suggest numerical comparative scale ratings may be significantly affected by rating-context effects Under these situations, researchers should consider using entity-based multiple store rating procedures since prior research indicates that, when compared to attribute-based rating scales, entity-based scales are less vulnerable to the types of context effects examined in this study (Teas and Wong 1992a). Because of the potential advantages associated with the NCS, there is a need for research that examines ways (0 reduce the vulnerability of the NCS to rating-context effects. For example, collecting NCS data orally by telephone interviews or personal interviews may decrease the effects of context store ratings by eliminating the respondents’ visual contact with the scale (‘Teas and Wong 1992a). Additionally, when using the NCS it may be useful to consider using an ‘anchor’ store as the first store for each respondent, This store, perhaps the most widely recog- nized store, may provide an effective anchor for the remaining ratings. Holding this store constant across individual respondents may have the Copyright © 2001. All Rights Reserved. Retail services image measurement 441 effect of holding the store context constant across the respondents even if the other stores being rated are not the same. For example, having cach respondent rate this ‘anchor store’ first on the numerical compara~ tive scale may result in stabilizing the context. The other stores are then rated from the perspective of this ‘anchor store’ Dr R. K. Teas Professor of Marketing College of Business Towa State University Ames 1A 50011 USA Notes 1 For example, as noted by Teas and Wong (19922), if different respondents rate different subsets of stores, how is an entity from one subset unambiguously com- pared with an entity from another subset, assuming the two subscts affected the ‘ovo entities’ ratings in different, unknown ways?” 2 In each of the group survey sessions, the students were randomly assigned to the three cells. The equivalence of the resulting three student samples was examined by comparing the samples in terms of percent female, age, and junior/senior clas- sification. No statistically significant differences were found (p < .03). 3 Perceptual mapping is a widely used technique in marketing in situations in which the marketing decision maker needs information about ‘I) the basic cogni- tive dimensions consumers use to evaluate “products” [or stores]... and 2) the relative positions of . . . products [or stores] on those dimensions’ (Hauser and Koppelman 1979). 4 In perceptual mapping applications, two-dimensional maps are often used. However, when more than two discriminant axes are necessary to exphin the mapping results, perceptual maps involving more than two dimensions are used. For examples of retail image maps in more than two dimensions, see Hauser and Koppelman (1979) A group (store) centroid indicated ‘the most typical [mean] location of an individ ual [observation] from a particular group, and a comparison of the group cen- ‘roids shows how far apart the groups are along the dimension being tested’ (Hair et al, 1992). In this study there are five groups of responses being compared; con- sequently, there ate five centroid values for each of the four discriminant function References Berry, L. L. (1969) ‘The components Evans, R. H. (1980) “The upgraded of department store image: a semantic differential: a further theoretical and empirical analysis’, Journal of the Market Research Journal of Retailing, 45: 3-20. ‘Society, 2: 143-7, Copyright © 2001. All Rights Reserved. 442. R. Kenneth Teas Gautschi, D. A. (1981) ‘Specification of patronage models for retail center choice’, Journal of Marketing Research, 18: 162-74. Golden, L. L., Albaum, G. and Zimmer, M. (1987) “The Numerical ‘Comparative Scale: an economical format for retail image measurement’, Journal of Resailing, 63: 393-410. Hair, J. F., Anderson, R. E., Tatham, R.'L. and Black, W. C. (1992) Multivariate Data Analysis, New York: Macmillan Hauser, J. R. and Koppelman, F. S. (1979) “Alternative perceptual ‘mapping techniques: relative accuracy and usefulness’, Journal of Marketing Research, 16: 495-506. Hirschman, E. C., Greenberg, B. and Robertson, D. H. (1978) ‘The intermarket reliability of retail image research: an empirical examination’, Journal of Retailing, 54: 3-12. Holbrook, M. B. (1983) ‘Using a structural mode! of halo effect 0 assess perceptual distortion due to affective overtones’, Journal ‘of Consumer Research, 10: 247-- 52. Jaffe, E. D. and Nebenzahl, I. D. (1984) ‘Alternative questionnaire format for country image studies’, Journal of Marketing Research, 2 463-71 Jolson, M. A. and Spath, W. F. (1973) “Understanding and fulfilling shoppers’ requirements’, Journal of Retailing, 49: 38-50. MeDongall, G. H. G. and Fry, J. N. (1974-5) ‘Combining two methods of image measurement’, Journal of Retailing, 50: 53-61, Marks, R. B. (1976) ‘Operationalizing the concept of store image’, Journal of Retailing, 52: 37-46. Menezes, D. and Elbert, N. F. (1979) “Alternative semantic scaling formats for measuring store image: an evaluation’, Journal of Marketing Research, 16: 80-7. Narayana, C. L. (1977) ‘Graphic positioning scale: an economical instrument for surveys’, Journal of Marketing Research, V4: 118-22. Nevin, J. R. and Houston, M. J (1980) ‘image as 2 component of attraction {0 intraurban shopping, areas’, Journal of Retailing, 56: 77-93, Pathak, D. S., Crissy, W. J. E. and Sweitzer, R. S. (1974-5) ‘Customer's image versus the retailer's anticipated image’, Journal of Retailing, 50: 21-8. Pessemier, E. A. (1980) ‘Store image and positioning’, Journal of Retailing, 36: 94-106, Schuman, H. and Presser, S. (1981) ‘Questions and Answers in Attitude ‘Surveys, New York: Academic Press. Sherman, S. J., Abim, K., Berman, L. and Lynn, S. (1978) ‘Contrast effects and their relationship to subsequent behavior’, Journal of Experimental Social Piychology, V4 340-50. Stem, D. E. and Noazin, S. (1985) “Effects of number of objects and scale positions on graphic position scale reliability’, in R. F. Lusch (ed.) 1985 AMA Educators’ Proceedings, Chicago: American Marketing Association, pp. 370-2. Swan, J. E. and Futrell, C. M. (1980) Increasing the efficiency of the retailer's image study’, Journal of the Academy of Marketing Science, 8 51-7. Teas, R. K. and Wong, J. K. (19924) “Item context and the stability of centity-based and attribute-based ‘multiattribute scaling methods’, Journal of Consumer Research, 18: 536-45. Teas, R. K. and Wong, J. K. (1992b) “A test of the effects of entity context on the stability of alternative rmultientty seating methods’, unpublished Iowa State University, Department of Marketing Working Paper. Wu, B. T. W. and Petroshius, S. M. (1987) “The halo effect in store image measurement’, Journal of the Academy of Marketing Science, 15: 44-51. Zelnio, R. N. and Gagon, J. P. (1981) “The construction and testing of an image questionnaire’, Journal of the Academy of Marketing Science, ‘Summer: 288-99, Copyright © 2001. All Rights Reserved.

You might also like