Research design and sample characteristics
Our empirical research is descriptive in nature. A field study was carried out over a period of 3 months in a Central European country by one of the authors, who approached various jewelry stores (from fashion-jewelry stores to high-end luxury stores) by means of cold canvassing. The reasons for the selection of jewelry sales are manifold: more complex and expensive products result in greater importance of the personal selling, which, in turn, requires greater explanation and consultancy. Since jewelries are typically not self-service products, a client must engage in a personal selling interaction. The stores ranged from small jewelry designers to luxurious international chains to cover a wide spectrum of different salespeople.
After the store managers provided their approval, the salespersons and potential clients were instructed to both wear a sociometric badge during their personal selling situation. To protect the participants’ privacy, the literal content of the sales conversation was not recorded. After the sales conversation, both parties completed a survey and provided some demographic data. The sample comprises 32 different dyads (consisting of 32 different salespeople and 32 different customers). The salespersons’ ages ranged from 21 to 56, with a mean of 34 years. Furthermore, 63% were female. The clients’ ages ranged from 18 to 71, with a mean of 36 years; 75% were female. The sales conversation lasted on average 8.5 min. After completing the survey, study participants were debriefed about the purpose of the study.
Measurement of nonverbal behaviors
Various methods exist to measure nonverbal behaviors. One commonly used approach codes specific behaviors manually based on an existing coding scheme; however, this approach is very time-consuming. Technological advances provide alternatives such as sensors, full body motion tracking systems, and eye trackers. Their applications, however, are primarily suited to laboratory-based settings. Quite recently, a wearable sensor package (sociometric badge) was developed, which “offers clear advantages over traditional methods since data is automatically collected by electronic sensors rather than humans” (Olguin-Olguin and Pentland 2008, p. 1).
To incorporate an innovative feature, this research utilizes these social sensors (infrared sensor, microphone, Bluetooth module, 3-axis accelerometer) for automatic data collection, which allows data to be exported in a spreadsheet format. Sociometric badges are similar in size to mobile phones and almost imperceptible for the user (Kim et al. 2012). They are small, unobtrusive, worn around the neck, and record data classified into several categories (i.e., body motion, speech, face-to-face interaction, and proximity). Each of these categories contains a number of different variables. In line with Olguin-Olguin, Gloor, and Pentland (2009), we focus on kinesics (variables: (a) posture activity, (b) posture (front/back), (c) posture (sideward lean left/right)), paralanguage (variables: (d) volume, (e) volume consistency, (f) pitch), and proxemics (variables: (g) total time of face-to-face interaction, and (h) total time of close proximity). The measurement units of the behavioral variables refer to technical details and not all of them are intuitively interpretable. Therefore, transformations are carried out for presentational convenience. Table 1 provides an overview of these variables and their interpretation.Footnote 1
Measurement of customer responses
We focus on a broad spectrum of customer response variables that are well-established in the sales and marketing literature including customers’ evaluations of salespeople (attitude), the product they promote (perceived product quality), liking of the store (intention to recommend), and actual purchase (yes/no). The charisma of the salesperson was measured on the nine-item scale by Khatri, Ng, and Lee (2001). The 7-point scales (with 1 = no approval and 7 = full approval) measuring attitude toward the salesperson (MacKenzie and Lutz 1989), perceived product quality (Grewal, Monroe, and Krishnan 1998), and intention to recommend the store (Maxham and Netemeyer 2002) comprised three items each. All scales show satisfactory psychometric properties (Cronbach’s α above .9).
Classification of nonverbal behaviors
Our hypotheses tried to simplify the research agenda by distinguishing between dynamic vs. static communication styles. Thus, as a first step, our analysis aims to classify salespeople based on their nonverbal messages. For classification purposes, hierarchical cluster analysis was conducted. Prior to the analysis, we performed a z-transformation on all nonverbal variables. We used Ward’s fusion criterion with squared Euclidean distance as a measure of dissimilarity to determine the number of clusters. The corresponding dendrogram suggested a two-cluster solution. A k-means cluster analysis fine-tuned this solution, as Table 2 (upper panel) presents the results. Entries in columns 2, 3, and 4 show averages per sample/cluster. To make these entries more easily interpreted, nonverbal behavioral variables have been transformed such that their domain falls between zero and one.Footnote 2 For descriptive purposes, the rightmost column of Table 2 provides p levels of a Kruskal Wallis test.Footnote 3 Two steps investigated and confirmed the reliability of the cluster solution: (a) split-sample (of salespeople) analysis of two randomly determined subsamples; (b) split-half (of nonverbal variables) analysis.
Analyzing the salespersons’ nonverbal cues, we conclude that salespeople belonging to cluster 1, labeled as “dynamic actives”, demonstrate enhanced posture activity and animated voice tone by varying their tone of voice in terms of loudness. In general, they speak louder than representatives of cluster 2 and show a higher pitch rate. This might be explained by the fact that this cluster is dominated by females, who generally speak on a higher frequency range than males (Peterson et al. 1995). Furthermore, representatives of cluster 1 are more oriented toward the client (i.e., forward lean). As pointed out by Arena, Pentland, and Price (2010), higher activity levels are indicative of a person’s excitement, whereas lethargic activity levels refer to disengagement.
On the contrary, representatives of cluster 2 are very static in their movement behaviors and do not vary their tone of voice and loudness while speaking, which is indicative of being monotone. Moreover, they stay in very close proximity to their clients (i.e., less than 1 m) and are frontally oriented toward them (they mainly stand vis-à-vis their client without moving apart). As a consequence, this cluster is labeled as “adhesive statics”.
All three kinds of nonverbal communication behaviors (i.e., kinesics—posture activity, paralanguage—front volume/consistency and pitch, and proxemics—face-to-face interaction and proximity) contribute to the classification task. A more detailed analysis indicates that front volume and proximity are of particular importance.
The same steps of cluster analysis investigated nonverbal variables of customers.Footnote 4 Interestingly, we find almost identical categorization results: a two-cluster solution with cluster sizes 26 and 6. Twenty-five salespeople are classified as “dynamic active” and all customers they served are also classified as such. Only one customer (pairing with an “adhesive-static”-salesperson) is classified as “dynamic active”. The middle panel of Table 2 demonstrates that customers executed quite similar nonverbal behaviors. This similarity can be explained by the fact that interaction partners show mirroring behaviors (Fatt 1998, Peterson 2005). Mirroring, or emulating nonverbal behaviors, occurs mainly unconsciously and establishes strong rapport among interaction partners. Moreover, research has revealed that “without the correct body language or paralinguistic cues customers are either dissatisfied or fail to develop the empathy with the provider” (Gabbott and Hogg 2000, p. 394).
The lower panel of Table 2 provides the profiling of the two clusters based on demographics and purchase behavior. As shown, the clusters do not differ largely for salespeople and customers.
Figure 1 shows the conceptual model; in statistical terms, this is a mediation model with dynamic vs. static communication style as the independent variable (taken from the preceding classification procedure), perceived charisma as the mediator, and customer responses as the dependent variables. Hayes’s (2013) PROCESS procedure, model 4, analyzed the data. Table 3 presents the statistical results.
Investigation of H1 (first panel of Table 3). There is a statistically significant impact of communication style on perceived charisma, i.e., the charisma of dynamic actives is on average perceived by 2.38 points (of a 7-point scale) better than the charisma of adhesive statics. The effect size of this relationship is considerable, i.e., .42. Thus, H1 is supported.
Investigation of H2 (second panel of Table 3, total effects column). There is a positive, statistically significant impact of communication style on all three response variables which are based on customers’ evaluations (attitude toward the salesperson—2.04 points; perceived product quality—1.49 points; intention to recommend the store—2.74 points). Missing data variability for the adhesive statics cluster prevented reliable estimates for the purchase behavior variable. On the whole, H2 is supported.
Investigation of H3 (second panel of Table 3, indirect effects column). There is a positive, statistically significant indirect (via perceived charisma) impact of communication style on all three response variables which are based on customers’ evaluations (attitude toward the salesperson—2.36 points; perceived product quality—1.90 points; intention to recommend the store—2.63 points). Table 3 also presents the corresponding bootstrap confidence intervals for a type I error of 5% based on 2000 bootstrap samples and substantial effect sizes according to Preacher and Kelley’s (2011) κ2. At the same time, direct effects of communication styles are consistently not significant and, therefore, we find full mediation for all evaluative response variables. Thus, H3 is supported.