The Consumer Motivation Scale: A detailed review of item generation, exploration, confirmation, and validation procedures

This data article offers a detailed description of analyses pertaining to the development of the Consumer Motivation Scale (CMS), from item generation and the extraction of factors, to confirmation of the factor structure and validation of the emergent dimensions. The established goal structure – consisting of the sub-goals Value for Money, Quality, Safety, Stimulation, Comfort, Ethics, and Social Acceptance – is shown to be related to a variety of consumption behaviors in different contexts and for different products, and should thereby prove useful in standard marketing research, as well as in the development of tailored marketing strategies, and the segmentation of consumer groups, settings, brands, and products.


Specifications
The scale was developed across a variety of contexts and products to ensure a generalizable goal structure Data source location Gothenburg, Sweden Data accessibility With the article ( þsupplementary file for further details)

Value of the data
Scale development requires a multitude of analyses to be performed, the results of which are often buried in supplementary files, if presented at all. By publishing all the relevant data in the present open access data article, we hope to increase transparency and offer researchers and practitioners alike a detailed overview of the procedures through which the Consumer Motivation Scale (CMS) was developed.
Researchers and practitioners interested in the CMS, or consumption goals in general, will find a detailed account of the structure and characteristics of influential consumption goals.
Researchers and students that wish to learn more about scale development may find this article useful as a practical and extensive example of the steps involved in the extraction, confirmation, and validation of psychological factors.

Data
This data article offers a detailed review of the development of the Consumer Motivation Scale (CMS; [4]). The objective of the research is to establish a structure of sub-goals that form a coherent and practical measurement scale which is: 1. Integrativeencompassing not only utilitarian, but also hedonic and normative goals; 2. Multi-dimensionaltaking potential sub-goals of the three master goals into account; 3. Context-sensitivemeasuring not only individual, but also situational variance; 4. Generalapplicable to a wide variety of consumption settings and products.
The scale development procedure follows Churchill's [12] paradigm for developing marketing constructs. The procedure consists of the following five steps: 1. Specifying the domain of the construct: The three master goals and their potential sub-goals are specified and described. 2. Item generation: Items were generated based on theories and scales related to the identified subgoals. 3. Establishing a factor structure: The goal structure was explored and purified on Sample 1 A, and confirmed on Sample 1B. 4. Convergent, discriminant, and construct validity: Additional data (Sample 2) was collected, thoroughly testing the convergent, discriminant, as well as construct validity of the emergent dimensions.
5. Criterion-related validity: Finally, more data (Sample 3) was collected to test criterion-related validity, showing that the dimensions explain choice between alternative products.

The domain of the construct
The Consumer Motivation Scale (CMS; [4]) builds on the goal-framework developed by Lindenberg and Steg [19], in which three overarching "master" goals are identified and described, namely: the gain goal ("to guard or improve one's resources"; [19], p. 119), the hedonic goal ("to feel better right Is popular among my friends -Note: Items without references were formulated based on our understanding of the theories related to the given goal. For references to these theories, please see the footnotes to Table 1. now"; [19], p. 119) and the normative goal ("to act appropriately"; [19], p. 119). The three master goals are assumed to consist of several sub-goals, that in turn are related to means and behaviors. In recent exploratory research by the authors of the present data article [3], the three higher-order goals were shown to be represented by multiple distinct sub-goals, as the gain goal emerged as two distinct dimensions, one dealing with thrift, and the other with financial security. Likewise, the normative goal emerged as two separate dimensions, one dealing with ideals and moral norms, and the other with social status and fitting in. It was concluded that. It was concluded that the hedonic master goal can likely be represented by multiple sub-goals as well, and that the dimensionality of the master goals should be examined further. To this end, an in-depth literature review was conducted with the purpose of identifying relevant theories and measures related to each of the three master goals. Based on this review, nine preliminary sub-goals were identified and described ( Table 1).

Item generation
Items were adapted or generated, based on a selection of established theories and scales related to the nine preliminary sub-goals. Generated items were formulated based on our understanding of the theories related to the given goal, while adapted items were based on the content of established scales (see Table 2 for references). To get information about the active goals, the scale is introduced with the statement "When I _, it is important that what I choose…" (where the blank is replaced by a suitable reference to the product under study, e.g. "shop for clothes"), followed by the list of items, represented as the continuation of that statement (e.g. "… is not too expensive"). Table 3.2 Factor extraction -The seven-factor structure.

Table 3.2 (continued )
Note: For increased readability, loadings 4.5 are bolded while non-significant cross-loadings are colored gray. Items in italics show items which were removed following scale purification. VfM ¼ Value for Money; Social Ac. ¼ Social Acceptance   Note: Non-equal models are bolded. Δ was calculated in comparison to the previous model.   Table 3.7) Table 3.7 The Consumer Motivation Scale (CMS).
Item # What matters the most to you when you _?

Value for Money VfM1
Reasonable price: the product should be reasonably priced VfM2 Not too expensive: the product should not be too expensive VfM3 Economy: the product should be economical VfM4 Value for money: I should get a lot for the price I pay VfM5 Not wasteful: the product should not be a waste of money

Quality Quality1
Quality: the product should be of consistent and high quality Quality2 First class: the product must be of the highest class Quality3 Well made: the product should be well-made or well-performed Quality4 Fulfills expectations: the product should fulfill even my highest requirements and expectations Quality5 Reliability: the product should be reliable (I should know what I get)

Safety Safety1
Security: the product should provide a prolonged and persistent feeling of security Safety2 Safe and secure: the product should feel safe and secure Safety3 Preparation: the product should make me well-prepared in case something unforeseen happens Safety4 Calm and safe: the product should make me feel calm and safe Safety5 Future needs: Needs that may arise in the future should be taken into consideration

Stimulation
Stimulation1 Exciting: the product should be exciting Stimulation2 Stimulating: the product should be stimulating Stimulation3 Avoid boredom: It is important that the product is not too boring or routine Stimulation4 Unique: the product should be unique (or give many unique experiences) Stimulation5 Interesting: the product should be interesting

Comfort Comfort1
Smoothness: the product should be smooth and comfortable Comfort2 Avoid inconvenience: the product should not be too inconvenient Comfort3 Avoid hassle: the product should not be too complicated or strenuous Comfort4 Pleasure: the product should be pleasant and agreeable

Ethics Ethics1
Not morally wrong: the product should not be morally wrong Ethics2 Principle: the product should not violate my principles Ethics3 Obligations: the product should be compatible with my personal and moral obligations Ethics4 Ideals and opinions: the product should be compatible with my ideals and opinions Ethics5 Good conscience: the product should give me a good conscience Social Ac. Social1 Friends' approval: the product should be approved by my friends Social2 Popularity: the product should be popular in my circle of friends Social3 Friends' expectations: the product should not go against my friends' expectations of me Social4 Good impression: the product should make a good impression on people who are important to me Social5 Liked: the product should be liked by people who are important to me Note: The dimension labels and the item # should not be visible when used in a questionnaire, and the items should be presented in randomized order. Social Ac. ¼ Social Acceptance 3. Establishing a factor structure

Data collection (Sample 1)
Nine-hundred eighty-seven respondents were recruited from a general population research panel at the University of Gothenburg, Sweden. The CMS was presented with the question "When I _, it is important that what I choose…" (where the blank was replaced by one of the following: "shop for Change to a hotel which has 4.5/5 user rating and lies in an area which recently has been selected "One of the summers' trendiest travel destination" by a panel of well-known fashion and travel magazines CSII Bearden et al., [6] VfM ¼ Value for Money; Social Ac. ¼ Social Acceptance food", "shop for clothes", "shop for something that is entertaining or amusing", "spend money on travel", or "look for housing"), followed by the list of items, represented as the continuation of that question (e.g. "… is not too expensive"). The participants were asked to rate the importance of the 63      items in their respective context, on a seven-point scale, ranging from 0 (not at all important) to 6 (extremely important). The sample was then randomly split into two halves, with exploratory analyses performed on the former (Sample 1 A), and confirmatory on the latter (Sample 1B).

Factor extraction
Principal component analysis (PCA) was conducted on Sample 1 A (N ¼ 496) to find the structure with the highest explained variance without signs of over-extraction. Over-extraction was defined as a structure with at least one factor made up of fewer than three main loading items [14], while a main-loading item was defined as an item with a factor loading of .5 or greater, and that does not have a cross-loading which amounts to at least .32 and 4 50% of the main loading [30].
The conditions for PCA were met in that KMO is larger than .6 (.923) and Bartlett's p is significant (p o.001; [18]). An oblique rotation was used since the dimensions are assumed to be naturally correlated.
Of the structures that meet the set criteria, the seven-factor structure is the structure with the highest explained variance (59.9%; Table 3.2.1). In comparison, the eight-factor structure has one factor that is made up of only two items, none of which load greater than .5.

Scale purification
The emergent structure was then purified by removing weak and cross-loaded items one by one, recalculating the communalities and factor loadings after each removal, with the following criteria being used for exclusion: 1. Insufficient communality (  [14,30].
One additional item was removed from factor III: Comfort3 "gives relaxation". The decision was based on content rather than communalities or cross-loadings. The other items in factor III are about stimulation and excitement, and so we opted to exclude this item to keep the factor relatively clean. The resulting 45-item scale consists of seven distinct factors, explaining a total of 65.7% of the variance.

Dimension labels
The seven emergent dimensions correspond to seven of the nine preliminary dimensions, and the labels were therefore applied accordingly. Factor I is entirely made up of items from the preliminary Safety dimension and the label was therefore retained. Likewise, factor II is made up of items from the Social Acceptance dimension, factor V of items from the Ethics dimension, and VI of items from the Value for Money dimension. For the remaining three factors, III, IV, and VII, one of the preliminary dimensions is clearly focal in terms of number of items as well as factor loadings, with a maximum of one or two items from another dimension. The preliminary labels were therefore retained for them as well. Only Function and Pleasure did not emerge as distinct factors.

Factor confirmation
Confirmatory factor analysis was performed on sample 1B (N¼491; Table 3.5) to confirm the emergent factor structure. A null model, in which all variables were assumed to be uncorrelated, was compared to four specified models, with increasing levels of separation between the dimensions. In model 1, all items load on a general factor, in model 2, all items load on factors representing the master goals: gain (Value for Money, Quality and Safety items), hedonic (Stimulation and Comfort), and normative (Ethics and Social Acceptance), in models 3A, 3B, and 3C, the master goals were split into subgoals one at a time: in 3A gain was split into the three sub-goals Value for Money, Quality, and Safety; in 3B hedonic was split into Stimulation and Comfort; and in 3C normative was split into Ethics and Social Acceptance. Finally, in model 4, all items load according to the PCA. A shorter version (model 4B) was then tested, with the purpose of achieving higher model fit and parsimony. In this model, items per dimension was reduced to a maximum of five, based on factors loadings as well as content (in general, we wanted strong loadings while maintaining varied content), resulting in a 34-item model.
Model fit improved significantly for each model (p o.001), with model 4 A and 4B both reaching satisfactory fit in terms of χ 2 /DF (o3) and RMSEA (o.10), while model 4B approach satisfactory CFI (.89; 4.9 is preferable, but these cut-offs are not clear-cut).

Invariance testing
The scale was tested for invariance on Sample 1B, first for factor loadings (metric invariance) and intercepts (scalar invariance), and then for factor means (structural invariance) and residuals (strict invariance). While the model should be invariant on the metric and scalar levels to allow for meaningful comparisons between contexts, it is assumed that the model is not invariant on the structural level, as the means are expected to vary from context to context, and we do not require our model to be invariant on the strict level, even though it would be preferable. As the name applies, strict invariance is simply not feasible for most models [32].
First, a baseline configural model was defined, in which the contexts were specified, but all parameters remained unconstrained across the contexts. Each subsequent model then builds on the previous model, but with added constraints for that level of measurement; factor loadings were constrained equal for the metric model, factor loadings and intercepts for the scalar model, factor loadings, intercepts, and means for the structural model, and finally, factor loadings, intercepts, means, and residuals for the strict model.
As a general criterion, a model was considered invariant if ΔCFI is not below À .01 in comparison to the previously accepted model [10]. If invariance does not hold for a model (i.e. ΔCFI is below À.01), then the test is repeated for each dimension, and if invariance does not hold for a dimension, the test is repeated for each item in that dimension, thus identifying the source of the non-equality. For partial invariance to hold, a dimension should have at least two invariant items. As the scalar model has a ΔCFI below À .01, invariance of intercepts is tested for each dimension on the scalar level of measurement (Table 3.6.2.).
As ΔCFI is below À .01 for Quality, Safety, Stimulation, and Comfort, scalar invariance is tested for each item in these dimensions (Table 3.6.3.).
All items are invariant for the Quality and Comfort dimensions, while Stimulation has at least two invariant items, thus meeting the criterion for partial invariance. Unfortunately, Safety only has one invariant item, and is therefore not sufficiently invariant on the scalar level. Since situational fluctuations in intercepts could influence the mean, situational differences in this dimension should therefore be interpreted carefully. However, two things should be noted: first, ΔCFI is not extreme, ranging from À .011 to À .014, and second, contexts that are very different from each other increase the likelihood that a model will fail to be invariant. The contexts in the present study were purposefully chosen as they represent different areas of consumption, and should therefore be considered a harsh test of invariance. For instance, when the housing context is dropped from the modelleaving food, clothes, entertainment, and travelall five Safety items achieve satisfactory invariance on the scalar level, with ΔCFI ranging from À .001 to À .005. Thus, situational differences in the Safety dimension should be interpreted carefully across contexts that are very different from each other (e.g. food vs. housing), but it is unlikely that this would be problematic in most scenarios.
For the final test, the partially invariant scalar model, in which the constraints for the non-equal intercepts were relaxed, was compared to the constrained structural (factor means) as well as strict (residuals) models (Table 3.6.4.).
As expected, ΔCFI is below À .01 for the structural model, which suggests that the means of the dimensions do indeed vary significantly across contexts. These results were replicated in a MANOVA: all dimensions vary significantly (F range from 6.3 to 39.64, all significant at p o.001, and η2 range from .05 to .26), except Ethics (F [4, 441] ¼ 1.32, p ¼.261, η2 ¼ .012). It is not within the scope of the present article to further test variations across contexts, please see [5], for a series of experiments testing the situational activation of the sub-goals across functional, hedonic, and social situations.
Invariance of residuals (strict invariance) is not supported across these contexts.

Finalized version of the Consumer Motivation Scale (CMS)
In the finalized 34-item version of the scale, the items are introduced by the question "What matters the most to you when you _?" (where the blank is replaced by a suitable reference to the product under study, e.g. "shop for clothes"), followed by the list of items, representing statements that answer the question. Note that the item wordings were modified slightly from previous versions, in order to improve clarity and allow it to be better tied to the context at hand. For instance, "is reasonably priced" was changed to "Reasonable price: the product should be reasonably priced" (where "the product" can be changed freely to suit the context, although this is optional). The participants then rate to what extent each statement is important to them in the given context, for instance rated on a six-point scale, from 0 (not at all important) to 5 (extremely important).

Data collection (Sample 2)
Two-hundred fifty-five respondents were recruited from a pool of voluntary research participants at the University of Gothenburg, Sweden. Participants were asked to what extent they search for different kinds of information before they decide where to go for vacation (rated on a six-point scale ranging between 0 [not at all] to 5 [to a very high degree]), and which of seven hypothetical travel package upgrades they prefer (rated on a five-point scale ranging from 1 [least preferred] to 5 [most preferred]). One information search behavior and one hypothetical travel package was formulated for each dimension of the CMS.
The final version of the CMS was used in this study (see Table 3.7), introduced with the question "What matters the most to you when you choose among vacation trips?" (note that "the product" part of the item wordings in Table 3.7 was changed to "the vacation trip" to suit the context), rated on a six-point scale, from 0 (not at all important) to 5 (extremely important).
The questionnaire also contained a selection of similar scales (from here on referred to as "reference scales"); one reference scale was selected for each dimension. For Value for Money and Quality, items from the PERVAL dimensions price and quality [29] were included and rated on the same scale as the CMS. For Safety, Schwartz's [25] security value type was selected, rated on a sixpoint scale ranging from 0 (not at all important) to 5 (extremely important). The sensation-seeking scale (SSS; [1]) was chosen as reference to Stimulation, rated on a six-point scale ranging from 0 (does not apply at all) to 5 (applies precisely). The restful experience dimension reported by Bello and Etzel [8] was used for Comfort, and was rated on the same scale as the CMS.
The universalism value type [25] was chosen for Ethics, rated on a six-point scale ranging from 0 (not at all important) to 5 (extremely important), and for Social Acceptance, consumer susceptibility to interpersonal influences (CSII; [6]) was chosen, rated on a six-point scale ranging from 0 (does not apply at all) to 5 (applies precisely).

Convergent validity
Convergent validity was tested on Sample 1 A by recalculating the PCA on the confirmed 34-item scale. The factor loadings should be at least 4 .5 (but preferably 4 .7; [18]), a criterion met by all items except one. The internal consistency of the dimensions is satisfactory, with Cronbach's alpha ranging from .81 to .88 (4 .7 is commonly regarded as acceptable).
Several measures were calculated on Sample 1B to test convergent validity (Table 4.2.2). According to Hair et al. [18], convergent validity is supported if average variance extracted (AVE) is greater than .5 and if composite reliability (CR) in turn is greater than AVE. These criteria were met for all dimensions except Quality (AVE¼.45). Low AVE may indicate low factor loadings, however, since four of the five factor loadings are significant on Sample 1A as well as 1B, this is not considered problematic for the dimension as a whole.
Bivariate correlations were calculated for each dimension of the CMS and its reference scale on Sample 2 (N ¼ 255). All dimensions correlate positively and significantly with their respective reference scale. Note that a single item from the security value type, namely "Family security", was chosen as reference scale for Safety due to insufficient Cronbach's alpha when the items were combined (.60). Note that as there are a few shared items between the CMS and the reference scales, whenever the CMS dimension and the given reference scale is correlated, the overlapping items were excluded from the CMS to avoid inflated correlations.

Discriminant validity
To test discriminant validity, the factor loadings and component correlations from the recalculated PCA were examined once more, this time to check for remaining cross-loadings and excessive overlap. There is only one remaining cross-loading, and correlations between components are not excessive (o .7).
Several measures were calculated on Sample 1B to test discriminant validity. According to Hair et al. [18], discriminant validity is supported if the average variance extracted (AVE) is greater than the maximum shared squared variance (MSV) as well as the average shared squared variance (ASV), criteria met by all dimensions.
To test discriminant validity on Sample 2, the "target" bivariate correlation coefficients in Table 4.2.3 were compared to the average correlation between a given dimension of the CMS and the six "unrelated" scales (i.e. the reference scales of the other dimensions). Fisher's r-to-Z transformation was calculated for significance testing.
The target correlations were significantly stronger than the average unrelated correlations for all dimensions except Value for Money and Safety. For Value for Money, the non-significant Z likely depends on the relatively weak correlation between this dimension and its reference scale (.36). Since Value for Money performs well on the previous tests of convergent as well as discriminant validity, this is not seen as problematic. For Safety, the negative Z value likely depends on the relatively strong correlation with the unrelated scales (.41). Excessive correlations between the dimensions were not observed in previous contexts, and the overlap is therefore concluded to be contextual. A high correlation between, for instance, Safety, Comfort, and Quality, may in fact be expected in a travel context, as few vacation goers would consider a vacation package that is unsafe and uncomfortable to be of high quality.

Construct validity
Construct validity was tested on Sample 2 by performing a series of regression analyses with the dimensions of the CMS as independent variables, and the 14 target constructs as dependent variables. For comparison, a parallel model was defined in which the CMS was replaced by the reference scales. All target regression coefficients for the CMS were statistically significant at p o.05. The reference model performed worse in this regard, as five coefficients out of 14 were non-significant.
Also, bivariate correlations were calculated for each dimension of the CMS/reference scale on the one hand, and the seven target constructs on the other. The coefficients of the CMS were then compared to the coefficients of the reference scales, using Fisher's r-to-Z transformation. The coefficients of the CMS were significantly stronger in three cases, and weaker in one case.
Note that due to limitations in size and format of the questionnaire, the reference scales could not always be included in full or exactly as intended by their authors (the stimulation-seeking scale, for instance, consists of 40 items alone). The comparison between the CMS and the bundle of reference scales should therefore not be considered a test of the reference scales themselves, but rather as a test of the principle of bundling different scales to represent a multi-dimensional measure. The difficulty in accomplishing this is in itself an advantage of the CMS.

Data collection (Sample 3)
Two-hundred fifty-six participants were recruited in a class room environment at the University of Gothenburg, Sweden. The participants were asked to make a hypothetical choice between a regular chocolate at the cost of 20 SEK (approx. €20), and a carbon-compensated "green" chocolate bar at the cost of 50 SEK (approx. €50).

Criterion-related validity
To test criterion-related validity, binary logistic regression was performed with purchase choice (regular vs. green chocolate) as the dependent variable and the seven dimensions of CMS as independent variables. The model correctly explained 74.2% of the choices, and Cox & Snell R2 as well as Nagelkerke R2 were satisfactory (. 30  For an increase of 1 unit in the Value for Money dimension, the likelihood of choosing green chocolate decreased by a factor of 3.19, while each increase of 1 in the Ethics dimension increased the likelihood by a factor of 3.32.