First, the MIX construct should include community elements from the AT framework because academic researchers and industry experts have discovered that omnichannel retail strategies can depend on successful community-building. Traditionally, community in retailing distinctly referred to the cities, towns, and neighborhoods that surrounded stores and shopping districts. While these cultural geographies remain crucial for retail operations, stores in the 21st century are once again becoming communities in their own right (Alexander and Cano, 2020), and they are launching quirky pop-up shops (e.g., Wayfair.com), larger than life flagship stores (e.g., Burberry Regent Street), stores within stores (e.g., Story at Macy's), and showrooms (e.g., Everlane). By pairing contemporary retail formats with mobile in-store experiences, stores can welcome consumers into their real-time, brand communities through behavioral targeting. Behavioral targeting is defined as context-aware and real-time mobile services that offer product recommendations, attractiveness cues, and promotions gathered from in-store sensor data, user behavior, and purchase histories (Van der Heijden, 2006; Krumm, 2011). Previous studies have supported that consumers can derive value from behavioral targeting when the outcome is consistent with their shopping behaviors and goals (Blom et al., 2017). For example, Brynjolfsson et al. (2013) demonstrate that successful behavioral targeting helps consumers to produce value and seek social validation by giving them opportunities to curate content, seek feedback and increase their product knowledge.

Furthermore, consumers can socially transcend retail servicescapes by using their mobile devices to incorporate friends and family into their in-store shopping activities (Houliez, 2010). For example, they can take a photo of themselves wearing apparel and upload it to Facebook, Instagram, and Twitter in order to review and share product experiences within their social networks (Morris et al., 2014). In this way, mobile in-store activities can lead consumers to heighten their interactivity and sociability when shopping in retail stores, which enables the co-creation of value with retailers and social networks during each contextualized exchange (Gummerus and Pihlström, 2011; Ström et al., 2014). According to AT, this interactivity and sociability can represent the shared interests of consumers (subjects) within retail stores (objects). In effect, consumers are utilizing their mobile devices as tools to build senses of community by mediating the subject–object relationship.

Secondly, recent findings and industry practices suggest that the MIX construct should capture aspects of the rules domain from AT. Due to the transformative characteristics of mobile devices and the formation of global brand communities, retail markets are operating according to a new set of rules. Dart and Lewis (2017) describe a seismic shift within retailing wherein mobile devices yield consumers more negotiating power, empower them to seek greater transparency from retailers, and give them access to nearly limitless inventories and on-demand services. These new rules are embodied in consumer trends such as webrooming and showrooming, which involves the assessment of products online prior to visiting a physical store and the purchasing products online from a competing retailer during or after a store visit (Jing, 2018).

To address these challenges, retailers must offer flexibility and selectively grant consumers permission through mobile devices to defy some conventional social norms within retail stores such as standing rather than skipping the checkout line, adhering to rigid store policies for couponing, and waiting for new products to arrive at the store (Shi et al., 2020). For example, Best Buy and Lowe's have invested significantly in mobile apps and technologies that tap into new, self-service opportunities presented by mobile media and devices (Pearson, 2017). Recent studies also demonstrate that mobile, self-service technologies in retail stores produce customer satisfaction, reduce wait-time perceptions, and create perceived value (Inman and Nikolova, 2017; Djelassi et al., 2018). Therefore, consumers are leveraging their mobile devices to mediate the subject–community relationship. Consumers (subjects) are effectively creating a new normal wherein mobile devices lead them to renegotiate their interactions with social agents such as important others and retail workers (community).

Thirdly, the MIX construct must include aspects of the division of labor domain from AT to mirror the new forms of consumer engagement in the omnichannel environment. Current research and industry practices illustrate that mobile devices have shifted the division of labor between frontline retail associates and store visitors. Traditionally, a salesperson or end-provider was the locus of activity and value creation within retail stores. Over the past decade, mobile devices and accelerating consumer demands have democratized store environments and increased consumers' direct involvement with the selling of goods and services (Lewis and Dart, 2014). RFID and Bluetooth connections especially have led consumers to increasingly assume some responsibilities from sales associates through mobile apps, such as wayfinding (e.g., REI), gathering product information (e.g., Sephora), and building gift lists (e.g. Target). Amazon Go, perhaps, is at the extreme end of these mobile in-store activities because consumers may optionally choose to never interact with sales associates (Grewal et al., 2017). Additionally, findings suggest that such participatory behaviors can strengthen customer relationships, enhance customer loyalty, and increase hedonic purchases (Juaneda-Ayensa et al., 2016; Mustak et al., 2016; So et al., 2016; Huang and Yang, 2018). Likewise, consumer participation is indicative of a gradual power-shift in which mobile devices are enabling consumers to collaborate with social agents (community) to manage and navigate retail stores (objects) in new ways. Specifically, consumers can use mobile devices to mediate the object–community relationship, between physical stores and social agents within retail servicescapes.

In the former sections (Development of the MIX Index and Step-by-Step Procedure), we defined the construct domain and postulated that three types of mobile-mediated activities can represent the MIX construct: community, rules, and division of labor. This observation was drawn from AT and substantiated with current m-commerce and omnichannel research (e.g., Grewal et al., 2017; Inman and Nikolova, 2017; Alexander and Cano, 2020; Shi et al., 2020). The next step calls for additional theoretical support to ascertain the dimensionality of the construct (MacKenzie et al., 2005). According to Jarvis et al. (2003), conceptual dimensionality means that a researcher must heed attention to the levels of abstraction within a construct. Indeed, the AT framework and literature support the notion that consumers and retailers create a mobile in-store experience from three types of activities. Nevertheless, the MIX construct should be unidimensional in nature, rather than having three unique subdimensions that give rise to a mobile in-store experience since activities and mediators are fluid.

AT asserts that, even though technological mediation appears to crosscut three domains, activities do not occur within a specific order and can be mixed or matched according to the needs and desires of the subject to produce an experience (Mickelsson, 2013; Hepi et al., 2017). This subjectivity and spillover are central tenets of the AT framework. By creating subdimensions, we would violate this basic principle and suggest that mobile-mediated activities fall into neat boxes. This index will represent a single trait: intentions to engage in omnichannel retailing through a mobile device. Morwitz and Munz (2021) define intentions as how much a consumer resolves or desires to act in a certain manner. As such, the unidimensional MIX construct is similar to other global scales that tap into behavioral and usage intentions including validated measures of prosocial behavior (Baumsteiger and Siegel, 2019), desires for new products (Lynn and Harris, 1997), social media use (Chintalapati and Daruri, 2017), online innovativeness (Goldsmith, 2001), and digital coupon redemption (Nayal and Pandey, 2020).

Mobile devices that consumers currently use in retail stores, such as tablets and smartphones, are similarly transformative as the tools described in the AT framework. According to Shankar et al. (2010), the key properties of mobile devices in the omnichannel retail environment include capabilities and qualities such as ultra-portability, location sensitivity, untetheredness, and personalization. These sociocultural and technical characteristics have led consumers to integrate mobile devices into their shopping trips long before mobile apps and social media were mainstream (Rigby, 2011). Likewise, retailers have to consider how they might leverage this mobile consumer behavior to produce commercial opportunities (Shankar et al., 2010; Dart and Lewis, 2017). A mobile in-store experience, therefore, includes shopping activities and retail services that are initiated by consumers and retailers within an omnichannel retailing environment.

Hence, we iteratively developed nine interaction intentions to capture elements of consumer engagement in omnichannel retailing through a mobile device by considering the major AT domains (community, rules, and division labor) and key properties of mobile devices (ultra-portability, location sensitivity, untetheredness, and personalization).

Based on this information, we developed three interaction intentions related to the notion of community during consumer engagement in omnichannel retailing. These activities are consistent with behavioral targeting and the personalization properties of mobile devices, which consumers can harness through affiliation with social networks and brands in order to enhance their immersion in retail servicescapes:

Based on these research insights, we developed three interaction intentions related to rules during consumer engagement in omnichannel retailing. These in-store shopping activities leverage the untetheredness and ultra-portability properties of mobile devices, which consumers can use to circumvent old rules and practice new rules during their interactions with relevant community members such as salespeople, shoppers, and important others:

As a result, we developed three interaction intentions pertaining to the division of labor during consumer engagement in omnichannel retaining. These in-store shopping activities build on the location sensitivity and untetheredness properties of mobile devices, which enable consumers to assume some employee responsibilities and co-create value within retail servicescapes, thereby impacting the division of labor as denizens in stores:

After generating the interaction intentions, we aimed to assess the content validity of the MIX index. To establish content validity, the full domain of the MIX construct should address the different ways that consumers can engage in omnichannel retailing through their mobile devices. Yet, mobile-mediated activities depend on the technological capabilities that software companies and retailers have hardwired into retail stores such as WiFi, Bluetooth, and RFID sensors. As a result, the MIX construct must be pragmatic and reflect the scope of mobile-mediated activities that are conceivable and available within a large number of brick-and-mortar stores.

To address these constraints, we conducted a semi-structured interview with the Vice President of Demandware (now Salesforce Commerce Cloud) who was the financial sponsor of this research. Demandware's cloud-based software supported retail operations for 350 well-known stores and 1,600 online stores worldwide, thus this executive was knowledgeable about the omnichannel capabilities that most retailers enable consumers to access while shopping. During the interview, this executive confirmed that the items described as mobile in-store shopping activities and retail services were currently supported by many department stores and specialty retailers. Thereafter, we pre-tested the index items with 10 undergraduate students enrolled in an upper-level survey and research methodology course for comprehension.

After generating the measurement items, we carefully assessed the relationship between the construct and its measures to prepare for data collection and scale purification. The rationale for taking this step is that many researchers underestimate the importance of choosing between a formative or reflective construct. Data simulations prove that when a researcher misspecifies the construct and its relationship with the measures, it leads to biased model estimates and inaccurate hypothesis testing (Jarvis et al., 2003). Within a formative construct, causality flows from the indicators to the latent variable in contrast to a reflective construct, in which the latent variable is hypothesized to impact or cause measured behaviors. Additionally, the indicators that underlie a formative construct are not presumed to be interchangeable or internally consistent. The indicators do not produce equivalent changes in the latent variable because they may represent a composite of mutually exclusive feelings, attitudes, or behaviors (Jarvis et al., 2003). Hence, the MIX index should be represented as a formative construct. Because the MIX index items are conceptualized as interaction intentions, these indicators are hypothesized to independently and cumulatively impact the latent variable.

To collect data for purifying the nine-item MIX index, we contracted with SurveyMonkey Audience to recruit consumers who had recently purchased fashion products and administer to them our online survey. Fashion products were defined as, “Items that you can wear such as clothing, shoes, and accessories, i.e., belts, ties, hats, scarves, leather goods, jewelry, etc.” We limited our survey to this context because fashion retailers have been recognized as innovators in the m-commerce and omnichannel space (Röcker, 2009; Hansen and Sia, 2015). Furthermore, fashion consumers interact heavily with products and salespeople in-store given the sensorial and symbolic aspects of fashion goods (Küchler and Miller, 2005). As a pre-screening question, respondents were asked, “How many fashion products did you purchase in the past 3 months?” Respondents were disqualified from participation if they purchased less than one fashion product in the past 3 months.

It was also imperative to recruit consumers with prior m-commerce and omnichannel shopping experience so that they could understand and respond accurately to the mobile in-store activities that would be presented to them in the survey (Hallikainen et al., 2019; Morwitz and Munz, 2021). Therefore, we pre-screened survey respondents based by asking them, “What technology do you use when shopping for fashion products?” Respondents were screened out if they did not choose 4 (almost always) or 5 (always) for either the “mobile tablet” or “mobile phone” options. After the pre-screening, 1,267 respondents from the United States and 951 respondents from Western Europe participated in the online survey (see Table 1), which constituted a 49.8% average response rate. When presented with the MIX index items, respondents were specifically directed to “Read the following statements about in-store mobile shopping services offered by fashion retailers and select an appropriate response.”

Mplus 7.4, a structural equation model (SEM) program, was used to specify and evaluate a formative measurement model (Diamantopoulos and Winklhofer, 2001). After specifying that the hypothesized nine-item MIX index was a formative construct, a Confirmatory Factor Analysis (CFA) was performed to test the reliability and validity of this index (Kline, 2011). After building the MIMIC model, the initial model fit was good (see Table 2): X² (6) = 27.79, RMSEA = 0.06, CFI = 0.99, TLI = 0.97, SRMR = 0.01 (Hu and Bentler, 1999; Chen et al., 2008). Next, the reliability of the nine-item MIX index was examined at the indicator level rather than at the construct level. The reliability cannot be assessed at the construct level because formative indicators are not presumed to be correlated (MacKenzie et al., 2011). As a result, interconstruct correlations are not substantively meaningful for a formative construct. The indicator reliabilities, alternatively, can be evaluated by looking at the Z-scores for the individual factor loadings (MacKenzie et al., 2011). The Z-scores for the formative indicators were >1.96, which suggested that the MIX index was reliable at the indicator level. A multiple regression equation generated the factor loadings, however, which increases the very minor risk of multicollinearity in a formative model. Multicollinearity is a sign of redundant indicators and prevents the researcher from effectively capturing the essential domains of a construct (Grewal et al., 2004; Dickinger and Stangl, 2013). To test for multicollinearity, bivariate correlations were calculated between the index items. The results showed that two of the items had correlations >0.70. Therefore, we individually regressed the problematic items onto the remaining seven items. The Variance Inflation Factor (VIF) for each regression statistic (<1.0) was less than the recommended 3.0 cutoff, which indicated multicollinearity was not present (Petter et al., 2007).

Next, we assessed the convergent validity of the MIX index and considered the extent to which its items account for variance in the latent (MIX) construct. The convergent validity of the survey items was assessed with the error term of the latent construct (Williams et al., 2003). The rationale is that the error term should be <0.50 assuming the items fully capture the construct meaning. The error term for the latent construct was 0.29, which confirmed the convergent validity for the nine-item MIX index. We also assessed the discriminant validity of the MIX index. Discriminant validity is the ability to distinguish a construct and its indicators from other constructs. In order to test the discriminant validity of the MIX index, it was critical to demonstrate that the construct was unique from similar forms of mobile use and communication. To that end, a measure of electronic word-of-mouth (eWOM) was added to the confirmatory factor analysis. According to recent studies, eWOM occurs in an omnichannel setting when consumers share and review opinions about products sold in physical stores through websites and mobile devices (Flavián et al., 2020; Grewal et al., 2020; Tyrväinen et al., 2020). Thus, it was important to distinguish this closely related but different phenomenon from the activities that comprise a mobile in-store experience.

eWOM was measured with the following question: “How do you tell others whether you like/dislike fashion products after you buy them?” Respondents were given six choices (text, email, tweet, blog, Facebook, or Pinterest), and they were asked to rate their eWOM frequency on a five-point, Likert scale (never to always). In the model, we allowed the MIX index items and eWOM to be freely correlated: X² (70) = 1991.38, RMSEA = 0.12, CFI = 0.86, TLI = 0.83, SRMR = 0.26. The latent covariance between the MIX index and eWOM items were then constrained to 1.0: X² (71) = 3017.23, RMSEA = 0.14, CFI = 0.78, TLI = 0.74, SRMR = 0.27. By performing a chi-square difference test, it is possible to see whether the constraint improves the model fit. In other words, if the unconstrained model fits the data significantly better than the constrained model with the latent constructs perfectly correlated, then discriminant validity exists (MacKenzie et al., 2011). After doing so, the chi-square difference test proved to be significant: X² (1) = 1025.85, p < 0. 001. Therefore, the MIX index had discriminant validity.

In order to further test the nine-item MIX index, survey data was collected from consumers in Western Europe 2 weeks later and retested the formative measurement model. Cross-validation can help establish whether sample characteristics have any effect on psychometric properties and determine if a scale is reliable within different populations (Churchill and Peter, 1984). Prior to commencing the online study in Western Europe, professional survey linguists translated and back-translated it from American English into British English, German, and French. Once again using SurveyMonkey Audience panels, online survey data was collected from respondents in the United Kingdom (n = 370), Germany (n = 362), and France (n = 219) (see Table 1). These countries collectively represent the largest share of the m-commerce market in Western Europe (Enberg, 2018). This second study, therefore, provided a rigorous test of the MIX index's reliability and validity.

Using Mplus 7.4 once again and drawing upon survey data from these European samples, the nine-item MIX index was placed into a Multiple Indicators Multiple Causes (MIMIC) model (Jöreskog and Goldberger, 1975). In order to replicate the prior study, we defined the first seven items from the index as formative while emitting paths to the remaining two items (Jarvis et al., 2003). After doing so, the initial model fit was good (see Table 2): X² (6) = 10.73, RMSEA = 0.03, CFI = 0.99, TLI = 0.99, SRMR = 0.01 (Hu and Bentler, 1999; Chen et al., 2008). Next, we examined reliability at the indicator level. The Z-scores for three of the nine items were <1.96. Upon reviewing the item factor loadings from Study 1, we found these same items previously had the weakest factor loadings and Z-scores. Their inability to reach significance when retesting the MIX index confirmed that these items were weaker than the others. Dropping an item from a reflective model does not alter the meaning of the construct. However, dropping an item from a formative model eliminates an essential piece of the construct, and by extension, necessitates strong theoretical justification (Diamantopoulos et al., 2008; Edwards, 2011). For these reasons, the weaker items were not dropped from the index. After testing the reliability of the items, VIF scores were calculated to assess multicollinearity. None of the VIF scores exceeded the 3.0 cutoff, which confirmed that multicollinearity was not an issue.

According to prior research (Williams et al., 2003), an error term <0.50 conveys an acceptable level of convergent validity. In Study 2, the error term for the latent construct was 0.33, supporting that the MIX index had convergent validity. In order to test its discriminant validity, we once again ran a confirmatory factor analysis and allowed the nine-item MIX index and the frequency measure of eWOM to be freely correlated: X² (70) = 1723.89, RMSEA = 0.11, CFI = 0.86, TLI = 0.83, SRMR = 0. 24. Subsequently, we constrained the latent covariance between both measures to 1.0: X² (71) = 2620.66, RMSEA = 0.14, CFI = 0.78, TLI = 0.74, SRMR = 0.26. The chi-square difference test proved to be significant: X² (1) = 896.77, p < 0. 001. Therefore, the nine-item MIX index had discriminant validity.