Serving Like an Organization: How Foodservice and Retail Workers Interpret Their Interactions With Customers

Occupational Processes

The most influential understanding of how customers impact workers comes from Hochschild (1983), who introduced the concept of emotional labor. This describes the moments when workers must suppress their feelings or emotions in social interaction in order to comply with company policies surrounding service. Hochschild argues that this emotional labor has negative effects on workers, either because workers must enact emotions they do not feel, or because they must realign their emotions in order to reduce cognitive dissonance. Researchers have specifically focused on “surface acting” in thinking about frontline food service and retail, where workers act in ways they do not feel, producing dissonance between one's internal emotional state and external activity (Groth & Grandey, 2012). Jobs in foodservice and retail are thought to, by nature, require more surface acting, and thus be less satisfying, since these jobs don't allow for as much complex problem solving or empathy with customers as other forms of service work (Singh & Glavin, 2017). Occupations that require specific types of interactions have been shown to be associated with feelings of inauthenticity, anger, depression, and burnout (Erickson & Wharton, 1997; Sloan, 2004). This line of research suggests that the negative impact of emotional labor is an inescapable occupational hazard, which comes with the territory of frontline customer service work in foodservice and retail. That is, any work that involves customer interaction of the type required in frontline foodservice and retail requires will force workers to engage in alienating and harmful emotional labor.

Beyond emotional labor, research also shows that workers who deal with customers are at greater risk for lower subjective well-being due to the conflict inherent in balancing the needs of the customer and the organization (Korczynski, 2008; Troyer et al., 2000). If the negative interactions that result from surface acting are a defining process for how workers understand their relationships with customers, then workers would describe their experiences as hypothesized below:

Research supports the notion that the emotional labor entailed in service work is associated with lower job satisfaction specifically. A number of studies have found that negative interactions with customers can have negative impacts on workers, and that this is the case for a wide variety of service jobs. For instance, negative customer interactions are strongly correlated with higher intentions to leaves the job, and lower job satisfaction (Grandey et al., 2004; Van Jaarsveld et al., 2010).

Intrinsic Satisfaction

A consistent refrain in the sociological research acknowledges that workers draw additional meaning from their interactions with customers. Particularly in healthcare, researchers have shown how helping patients sometimes leads healthcare workers to a distinct sense of satisfaction with their work (DiCicco-Bloom & DiCicco-Bloom, 2019; Lopez, 2006). Even outside of care work, however, a small body of research suggests that workers continue to find deeper meaning in their interactions with customers (Endrissat et al., 2015; Misra & Walters, 2016; Reich & Bearman, 2018). Although it is not the main theoretical contribution of their studies, both Hochschild (1983) and Leidner (1993) acknowledge that workers often describe customer interactions as the most satisfying aspects of their jobs. Tolich (1993) studies grocery store workers, and particularly cashiers, finding that these workers sometimes feel connected to and closer with the general community due to their interactions with customers. Tolich goes so far as to argue that, rather than alienating, service work has the potential to be liberating, and intrinsically satisfying.

If this, instead, is a defining way workers conceptualize their customers, then the following hypothesis would hold:

Organizational Processes

Another ongoing line of research has highlighted the extent to which the employer shapes customer service interactions. Korczynski (2008) argues that the organization's role is fundamental to defining how worker's experience their customers. Service work, Korczynski argues, provides an inherent conflict where rational business and irrational customers have different needs. How organizations attempt to resolve this conflict is consequential.

Some studies deal specifically with how these strategies impact emotional labor. In a study of caregivers at different nursing homes, Lopez (2006) highlights the power of these organizational rules. Lopez shows that organizations can, under the right conditions, provide an avenue for emotional connection with their clients, just as they can provide an avenue for alienating emotional labor. An additional labor process is managerial support. Through debriefing after a difficult interaction, workers may be less likely to experience these interactions as emotionally draining (DiCicco-Bloom & DiCicco-Bloom, 2019). Additionally, some organizational structures which emphasize customer sovereignty may create a system where workers are more likely to experience abuse from customers and thus will have to conduct more emotional labor (Korczynski & Evans, 2013).

In addition to management strategies regarding emotional labor, organizational strategies surrounding the relationship between workers, customers and managers have also been shown to play a role in what workers take away from their customers. For instance, as Leidner (1993) notes, the complex and shifting triangle between workers, customers, and managers plays a role in shaping the quality of interactions. A manager's decision to routinize interactions through the use of scripts for Leidner's McDonald's workers can cut both ways, making it difficult to connect with customers or be authentic, but also providing a useful script when things go wrong. In a broader study of nursing homes, Vermeerbergen et al. (2021) also shows the power managers have in shaping service interactions and what workers take away from their customers. Even the boundaries between different service providers within the organization can create conflict and could alter how workers interpret their interactions with customers (Subramanian & Suquet, 2018). In frontline service work, these relationships are complicated by the presence of multiple levels of middle managers, often paid only a small amount more than frontline workers yet given the responsibilities of a manager (Bolton & Houlihan, 2010).

Other researchers highlight organizational attempts to construct how workers evaluate their customers as an additional labor process in service work that can have disparate effects on workers (Grant et al., 2009). Those workers who buy into the organizational culture, may get additional satisfaction out of interactions with customers. However, for workers who feel they must act as if they understand their customers only in the organizationally sanctioned way, even when they do not, may face additional dissatisfaction due to their forced inauthenticity. Managers also make use of customer feedback or other forms of surveillance as an additional form of control with respect to customer interactions (Brown & Korczynski, 2010; Fuller & Smith, 1991).

However, a question remains in this literature. What, if any, opinions do the workers have about how these organizational processes may alter customer service? These organizational processes may operate undetected by workers, altering the underlying nature of customer interactions. On the other hand, workers may also actively implicate organizational strategies when describing their interactions with customers. Two additional hypotheses stemming from these organizational processes are:

Implications for Job Ratings

This intervention is also informed by research on job satisfaction. Multiple studies have found that workers draw meaning from their work in ways that are sometimes unexpected and difficult to explain (Misra & Walters, 2016; Reich & Bearman, 2018; Storer & Reich, 2019). Even in jobs that provide low pay, benefits, and offer few rewards, for example, workers still find reasons to be satisfied with their jobs (Deery et al., 2019). While the objective qualities of a job in terms of pay, benefits or insecurity may or may not predict subjective well-being (Brown et al., 2012; Schneider & Harknett, 2019a; Warhurst, 2021), it can play a useful role in understanding how workers evaluate their job quality.

Research has tied both organizational and occupational aspects of customer interaction to job satisfaction. From an occupational perspective, research shows that jobs which require more surface acting are correlated with worse outcomes, including lower job satisfaction (Bhave & Glomb, 2016; Singh & Glavin, 2017). However, research shows that organizational support can lead to higher job satisfaction. In a survey of 338 retail workers, Duke et al. (2009), for instance, finds that workers who have higher levels of perceived support from the organization report higher job satisfaction when faced with the need to do emotion work. On the other hand, a lack of quality training on how to deal with customers may leave workers feeling incompetent, dissatisfied, and lead to turnover Sallaz (2015).

This paper analyses how different orientations towards customers are associated with job satisfaction. That is, are more extreme responses to job satisfaction, positive or negative, associated with certain conceptualizations of customers?

If workers who see customers as an occupational benefit or hazard are deeply impacted by this aspect of the job, then:

However, if supportive or unsupportive organizations are associated with stronger attitudes, then:

Advantages of Glassdoor

The Glassdoor dataset has significant advantages. First, it allows for free text responses in which workers can decide what to post based on what they feel are important aspects of the job. With respect to orientations towards customers, this dataset allows for an approach where workers integrate customers into discussions of their job quality on their own, without any researcher intervention. This approach, while more difficult to implement, provides insight on a large scale into how workers view their own experiences. This provides a stark contrast to interview methods or traditional surveys. The mechanisms by which an interviewer coaxes specific stories out of a worker, and translates those into a narrative, leaves the possibility that the interviewer may be playing an undue role in producing their data. As Swidler (2003) shows, individuals internal thought processes are relatively scattered, and may therefore be somewhat suggestible. Interviews with service-sector workers which ask directly about customers may lead respondents to produce a narrative in which customers, or certain types of interactions with customers, are more central than they otherwise would be. These methods are not mutually exclusive. By producing an account of the ways that workers make sense of their customers without any prompting, an analysis of the Glassdoor data can open new pathways to further questioning of workers.

Additionally, Glassdoor data is employer-identified, contains occupation data, and is voluminous. Quantitative research on customers often takes advantage of small samples from a single organization, or national samples which do not indicate organizations at all. Since comparisons between employers typically show that organizational practices and occupational characteristics can alter how frontline workers interact with customers (Korczynski & Evans, 2013; Lopez, 2006), accounting for unobserved organizational differences in how customer service interactions are managed is essential for identifying the relationship between how workers make sense of their customers and job satisfaction. Employer-identified data and employer fixed effects can take these unobserved differences into account (e.g., Schneider & Harknett, 2019a). Similarly, controlling for occupation is necessary since this may lead to variation in how customer interactions are structured (Gutek et al., 2000; Korczynski, 2009; Leidner, 1993).

It should be noted that this is a major advantage of this dataset, since few studies of customer service interactions span multiple organizations. Moreover, the large number of reviews included in this study presents a significant increase in sample size over other quantitative studies of customer interactions in the service sector.

Limitations of Glassdoor

The Glassdoor dataset is not without limitations. Most importantly, posts on Glassdoor are not randomly sampled. Individuals must spontaneously decide to post a review on the site, so the data-generating mechanism is opaque. Workers may post reviews more often after a negative or positive event, for example, leading to skewed results. The convenience sample does not allow for a definitive statement about whether workers in the overall population hold the same attitudes about customers. They do, however, provide some suggestive evidence about where more research should be done. In addition, comparisons within the sample may be somewhat less fraught, under the assumption that the same data generating mechanism operates for all workers who post on Glassdoor, regardless of industry, organization, or occupation.

In addition, reviewers may be lying about specific aspects of their demographic characteristics in order to avoid identification. Fortunately, since all of the potentially identifiable aspects surrounding Glassdoor reviews are optional, there is less incentive to fabricate information. Instead, these reviewers can simply leave fields blank. However, employers themselves could fabricate reviews in order to create a positive image of the company.

Another drawback of the data is that certain important demographic characteristics, such as age, race, and gender, are not included in the data. This is particularly troubling with respect to service work, since researchers have identified important ways in which emotional labor differs by each of these characteristics (Polletta & Tufail, 2016).

Finally, frontline workers may hold views about their relationships with customers that they, for whatever reason, choose not to express in the Glassdoor reviews. Since this is digital trace data, there is no possibility to follow up with a specific Glassdoor reviewer about how they are discussing their customer. To contend with these limitations, this paper also draws on data from an original survey, which will be described in more detail following the Glassdoor analysis.

Research Design

This paper uses the Glassdoor data in two ways. First, using qualitative text analysis, I performed an iterative coding operation, randomly selecting 500 Pros and 500 Cons reviews that mention customers. Using an inductive coding scheme, I first developed over 100 codes defining the different ways in which workers discuss their customers. Second, I reread and recoded the initial codes based on overall patterns in the reviews, consolidating codes where the attitudes were similar, but the phrasing or specifics of the complaint were different. This paper presents my findings in the following section and discusses them with reference to the orientation Hypotheses 1a–e, described above.

In a second set of analyses, this paper measures the extent to which these orientations are associated with job satisfaction. Using the 1,000 qualitatively coded reviews, I trained a machine learning algorithm to code the remaining 14,000 reviews. I then run regressions on job satisfaction for reviews mentioning customers in the Pros or Cons sections of the written job evaluation. I evaluate the extent to which these regressions support occupational or organizational primacy, Hypotheses 2a and 2b.

In a final set of analyses, I extend the Glassdoor research, drawing on an original survey of 849 workers from the same 10 retail companies. This survey asks questions derived from the analysis of the Glassdoor data, while also including demographic information, as well as an additional outcome of interest, turnover intentions.

Occupational Orientation

The most common descriptions of customers in reviews discussed how brief negative or positive interactions with customers affected the reviewer's job. Here customers are typically described as “rude” or “nice.” These reviews are distinguished from others by how they do not discuss these interactions as impacted by any other social process and are instead simply part of the job. I describe these conceptualizations of customers as Occupational, in which customers, good or bad, are thought to be an unavoidable part of service work.

Occupational Hazard. In this code, workers describe customers based on overall negative traits that speak mainly to the personality characteristics of customers themselves. As one reviewer discusses in the Cons section, “[s]ometimes customers make messes and are rude.” Similarly, another reviewer didn't like “dealing with terrible customers.”

Sometimes, although rarely, these negative reviews explicitly implicate the difficulties of emotion work that are tied to rude customers. One reviewer reports disliking “[r]ude customers”, and that they “can't talk back.” Another describes a con as “puttin on a show for customers.” Even in their discussion of emotion work, however, these reviews focus primarily on the customers themselves. As one reviewer writes, “[p]utting on your overly happy face all the time can be tiring. Some customers need to be brought down to earth.” In these reviews, there is a sense that it is the customers who are at fault for producing a negative interaction. Another reviewer writes, “some customers can get really demanding and needy, but that's true for all customer service jobs.” In short, rude customers are simply an occupational hazard, and a notable negative characteristic of the job.

Occupational Benefit. These types of reviews, in which customers are simply part of the job, are not uniformly negative. On the positive side, one worker writes “most of the customers are really nice.” These answers are often quite short, simply stating in the Pros “the customers,” “interactions with customers,” or “modest, nice customers.” One reviewer writes, “I loved my interactions with customers. They always kept a smile on my face.”

As in the occupational hazard coding, these positive reviews do not implicate any deeper feelings surrounding the customers, or any organizational features that impact this relationship. These workers simply seem to focus on the interactions with “friendly,” nice customers when summarizing their customer service interactions.

Intrinsic Value Orientation

Two additional codes, both in the Pros section of the text, emerged from comments discussing the intrinsic value workers received from interactions with customers at work. These reviews go beyond saying customers were nice, or talking about their personalities, and instead mentioned how they both enjoyed developing relationships with customers and helping people.

Helping. Some workers discuss an intrinsic satisfaction they feel in helping customers. One reviewer states,

“I loved working in an environment where people told me that I made their day better just by being cheerful. It is also quite enjoyable to make things like ice cream cones … Not because it is a complicated task, but because you made something for someone who then gets to enjoy what you made for them.”

Another worker echoes these sentiments, writing “I loved sharing my knowledge with my customers, it gave me a lot of self-satisfaction to see my customers walk away with a smile.” In this code, workers are not merely satisfied with the positive interactions they have with their customers. Instead, they find a deeper level of satisfaction in helping the customer, in making their day just a little bit better.

Relationships. Another positive orientation towards customers focuses on the social relationships workers have with their customers. These reviewers often write about “their” customers, such as in the statement “I love my customers.” They also describe how “regular customers make interaction fun.” One reviewer writes that it is “fun to connect with the customers,” and another reviewer goes further, describing their regular CVS customers as “like family to us.” Reviewers in this category often make a statement regarding “repeat customers.”

Organizational Orientation

The final set of reviews implicate the organization while discussing customers.

Supportive Organization. Some workers indicate a concern regarding whether it seems as though their employer cares about customers. One positive reviewer states “the company wows on customer service.” Another reviewer writes that the organization creates “a family like atmosphere that cares about the customers." This also means being given leeway to help customers. One reviewer gives a positive review of the company because “You have the power to do what you feel is necessary to please the customer.”

Others appreciate the organizational rules which create a positive relationship between customers and workers. One reviewer notes that their company has “incentives for high customer satisfaction.” Blending customers and workers together was also common, such as in this review: “the company works hard to create an experience for partners and customers where everyone feels recognized.” Workers in this category were also concerned with the quality of the product they were offering. One reviewer sees it as a positive that there is a “strong company focus on keeping prices down to save customers money.” Putting these concepts together, one reviewer writes in the Pros that “It is a company that values not only customers, but also the employees who serve those customers.”

Beyond organizational incentives and institutionalized support, workers were also invested in whether others cared about customers. One reviewer notes, “most employed have a genuine want to be part of an outstanding customer experience.” These reviews also denote positive relationships with managers, for instance appreciating when managers take their side in an argument with customers. As a pro, one reviewer states “I got yelled at by customers, but managers always had my back.”

Unsupportive Organization. While a number of workers emphasized positive aspects of how their company facilitated customer interactions, another set of workers tied their negative reviews to how the organization hindered positive customer relations. One of the problems identified in this regard is understaffing. As one cashier succinctly puts it, “too many customers not enough registers.” In this situation, workers are negatively impacted by the social stress of a long line. Another reviewer identifies the relationship between rude customers and understaffing - “When labor is cut to bare minimum, wait times increase for customers which can make impatient customers more likely to be rude.” Another reviewer echoes these feelings, saying “often we were understaffed and there would be a line out the door, this was stressful because then the customers would become personally upset with you even when you were working your hardest.” Others were bothered by their inability to help customers, and the repercussions they might face from managers if they did so. “Customer service was discouraged by heavy workloads and threats for being found in another department or stopping current tasks while helping a customer.” It is worth noting that these experiences are not always viewed as an organizational issue. For instance, one reviewer with an occupational hazard orientation mentions as a con: “People not understanding the volume of customers our store serves.” From that perspective it is not a question of understaffing, but of undue expectations from the customer.

Some were disappointed by the organizations seeming lack of care for customers. One reviewer writes:

“The customer service/ taking care of our people culture that we had all grown up with disappeared. In it's place was a culture of making a number just for the sake of making a number. The company's new found focus on top line sales, which has nothing to do with being profitable and creating shareholder value, only shows that this is still the case. In addition to the stores being grossly understaffed, as voiced not only by our managers, but also by our customers the company has made it extremely difficult to order the product that is needed to take care of the customers.”

Workers also were put off when they felt that the company took the customers side over the workers. “Upper management and corporate took customer complaints too seriously.” Another reviewer states “[The company] has this issue with allowing customers (not guests because they don't deserve that title) to act like animals and punish us for standing up for ourselves or explaining the outcome to them.”

Finally, reviewers were sometimes unhappy with how managers treated them in front of customers. One reviewer writes, “You get yelled at for anything and everything and in front of customers.” The implication from these reviewers is that the workers are concerned with how customers see them. When managers discipline workers in front of customers, this adds another layer to negative attitudes.

Other

Customer Service Experience. A final group of workers see their interactions as a way to gain customer service skills. These reviewers believe that these experiences will be helpful in other jobs down the line. One reviewer describes the positive aspect of their job as “A great company to learn customer service,” and another states that a job will “teach kids about how to approach customers and be personable.”

The table below details the frequency of each customer orientation in the subsample of workers. The frequencies of these codes show that two sets of orientations - occupational (hazard and benefit) and organizational (supportive and unsupportive), were the most frequent in the reviews (Table 1).

This qualitative analysis shows that there is significant variation in the ways that workers incorporate their customers into job evaluations. On the one hand, many workers do indeed see their interactions with customers as primarily negative, yet simply as an occupational hazard, as predicted in Hypothesis 1a: Occupational Hazard. For these individuals, the stress of dealing with customers is an inescapable part of customer service work. Some workers even make the connection between negative interactions and alienation from the self, describing how tiring it can be to be forced to act in a way in which you do not feel. Indeed, of the reviews which put customers in the negative parts of a job quality evaluation, the majority of workers describe their customers in this way.

However, this is not the only interpretation of customers. Thirty five percent of the reviews mentioning customers in a positive light discussed enjoying interacting with nice customers, as predicted in Hypothesis 1b: Occupational Benefit. Nearly thirty percent of the positive reviews of customers are predicted by Hypothesis 1c: Intrinsic Value, going beyond a simple positive description and identifying a deeper meaning that workers draw from interactions with customers. These workers derive significant intrinsic value from the act of helping customers, or their long-term relationships with customers.

Additionally, while previous research suggests that organizations play a role in shaping customer service interactions, these findings show that workers often implicate management and organizational strategies in discussing the quality of their interactions with customers. This analysis finds support for both Hypothesis 1d: Unsupportive Organization and Hypothesis 1e: Supportive Organization. Over 35 percent of the negative reviews of customers clearly implicate organizational strategies, such as understaffing, managerial miscommunication, or a focus on customer service statistics, when discussing how customer interactions detract from job quality. On the other hand, some workers appreciate when they notice their employer making an effort to improve interactions with customers. Roughly fifteen percent of the positive reviews mentioning customers mention how managerial support or organizational latitude in dealing with customers allows workers to feel good about how they are treating their customers.

The remainder of this paper focuses specifically on understanding the impact of this relatively understudied aspect of frontline workers interpretations of their customers. That is, how important is it to frontline workers that they believe their organization is supportive of customers and customer interactions? To do so, this paper use two strategies. First, it analyzes the relationship between the unprompted expression of an organizational orientation in a Glassdoor review on job quality evaluations. Second, the paper utilizes a survey in order to extend the investigation of the organizational orientation by analyzing the association between respondents’ opinions regarding how customers are treated by their employer, and two outcomes: job satisfaction and turnover intentions.

Dependent Variable

Job Rating. This variable is rated from one to five and is an employee's rating of the job. A Glassdoor reviewer must always supply a job rating. When posting on Glassdoor, under the header “Company,” workers add an anonymous company review.

Independent Variable

Customer Orientation. In two separate sections, workers are asked to share the upsides or downsides to working at their employer, under the headers of Pros and Cons. Orientations towards customers are indicated using a series of binary codes for either “occupational” or “organizational” orientations developed in the qualitative analysis above. Depending on whether the orientation appears in the Pros or Cons section of the data, the orientations coded are Occupational Hazard, Occupational Benefit, Supportive Organization, or Unsupportive Organization. These codes are extrapolated to the whole sample of reviews mentioning customers using the qualitatively coded subsample as a training dataset for a supervised machine learning model. For a detailed description of the machine learning method implemented, see Appendix A. The choice to code for these two orientations, and to leave out codes for intrinsic meaning or customer service experience, were based on two factors. First, organizational and occupational codes represented the two most common types of orientations and therefore represent a larger sample group to train a machine learning model. Second, the quantitative analysis is focused specifically on understanding more about how attitudes about the organizations support or lack thereof as an understudied aspect of the customer service interaction.

Controls

I include a control for character count, the number of characters in the Pros or Cons section of a written review. Length of review is included as it may serves as an indicator of how strongly a reviewer feels, where longer reviews may indicate a stronger attitude. Longer reviews, for instance, may cover more positive or negative aspects of the job, or may include a more detailed description of a specific pro or con. Counts are included from both the Pros and the Cons sections in each model.

In addition, I include a control for managerial level. Each company structures its employment hierarchy in different ways. Using information from each companies’ descriptions of job titles on their websites, I code job titles based on three hierarchical levels. Front-line non-managerial workers, including cashiers or sales associates, are coded as a one. Middle managers, including department managers, “head” cashiers, or team leaders, are coded as a two. Top level store managers are coded as a three.

I also include controls for whether a person worked on the job full time, worked on the job for less than a year, or worked in an urban area (a population of over 40,000 people).

Finally, I include a variety of fixed effects. Company-level fixed effects are included, since they play an important role in controlling for unobserved heterogeneity between employers in how service interactions are managed. Year-level fixed effects are also included. State-level fixed effects are also included, since the quality of customer interactions varies in unobserved ways due to the norms of a specific region.

Analytic Strategy

The analysis uses two subsamples of the Glassdoor data. The first subsample consists of reviews that mention “customer”, “guest”, or the frequent misspelling “costumer,” as well as their plurals, in the Cons section of the written text review. The second subsample selects on the same mentions of customers in the Pros section. I use listwise deletion, rather than multiple imputation for item non-response, due to the small number of controls and missingness around two key variables that would be difficult to impute - location and job title. For a more detailed discussion of missingness and listwise deletion, see Appendix B.

In the current, as well as the proceeding analysis, this paper presents the results of linear regressions based on the recommendation from Angrist and Pischke (2009), who find that linear regression performs as well as ordinal regressions while producing estimates that are more easily interpretable and subject to less restrictive assumptions. Standard errors in each regression model are clustered by company, since reviews were drawn specifically from separate companies. Finally, standard errors are adjusted using bootstrapping, in order to account for the predictive uncertainty in the machine learning model.

Descriptive Statistics

Below are the summary statistics for the variables included in the model. For a more detailed discussion of the Glassdoor data, see Appendix B. Since two separate models are analyzed for reviews, one that mention customers in the Pros section and another that mentions customers in the Cons section, descriptive statistics are included separately for each subset (Table 2).

The job quality ratings that mention customers in the Cons are slightly lower overall, 3.17, than the ratings that mention customers among the Pros, 3.31. The Cons reviews are more likely to mention both Organizational and Occupational concerns than are the Pros reviews. This makes sense, since the Pros reviews also included a significant number of intrinsic value and customer service experience orientations. Organizational concerns are also notably less common than occupational concerns in both models, an additional trait that is mirrored in the qualitative subsample. The reviews are written mostly by non-managers, making up about 75 percent of reviews mentioning customers in the Cons and 70 percent of reviews mentioning customers in the Pros.

Results

A summary of the results of the quantitative analysis are presented below. For a discussion of the tables associated with the model below, see Appendix C (Figure 1).

The figure demonstrates the average marginal effects for workers who hold these four separate orientations. For reviewers who express feelings of having an unsupportive organization when discussing customers, they give a rating of their job on Glassdoor a 2.8. Workers who discuss customers as an occupational benefit or hazard both rate their jobs as a 3.2 on Glassdoor. Finally, workers who discuss a supportive organization rate their jobs as roughly a 3.4. The standard deviation of the Glassdoor rating for the Pros analysis is 1.20 and for the Cons analysis is 1.21. So, the average marginal effect of discussing customers as part of an unsupportive organization is a decrease of about one third of a standard deviation, and the effect of discussing customers as part of a supportive organization is an increase of about .17 standard deviations.

With respect to the hypotheses developed above, this analysis provides significant evidence for Hypothesis 2b: Organizational Primacy. Workers who see their organization as unsupportive, hindering a positive social interaction with their customers, are significantly less happy with their jobs, while worker who see their company as supportive are the most satisfied. On the other hand, there is not support for Hypothesis 2a: Occupational Primacy. Indeed, when workers see negative interactions with customers as simply an occupational hazard or as an occupational benefit, their job ratings are the same. This is true even after including a host of controls, as well as company fixed effects.

These results point to two separate, but interrelated, underlying distinctions. First, there is a difference in the way that workers evaluate customer interactions - either as a dyadic interaction between a customer and a worker that would occur in any customer service job, or as the result of an organizational process. Second, some workers hold an underlying attitude about whether the company does a good or bad job in the way it deals with customers. This attitude has been an overlooked aspect of job satisfaction for frontline workers. The Glassdoor data suggests that this attitude is particularly salient for job ratings, more so than positive or negative views of a dyadic interaction with customers.

In order to extend the analysis of the organizational orientation, this paper draws on a survey which can more fully interrogate the underlying attitude: agreement or disagreement with how the company treats it customers. This method too is not without its drawbacks, since frontline workers may be providing answers to questions that are not typically part of their evaluative framework. However, given that this viewpoint has been identified in the unprompted Glassdoor data, the survey research is meant to develop a more well-rounded understanding of the relationship between this attitude and outcomes such as job satisfaction and turnover intentions, with the continued caveat that some frontline workers may not typically conceptualize their relationships with customers in this way.

Data

The survey was marketed to workers who indicated they were employed by the same 10 companies on Facebook.com from November 7th to November 12th, 2020. In order to be included in the sampling frame, individuals must be on Facebook and indicate that they work at one of the ten companies. This sample was collected using Facebook Advertising, in which workers who indicate that they are employed by these companies are served an advertisement on their news feed to fill out a survey for a chance to win a $500 Amazon gift card. This strategy has been used by the Shift project to analyze frontline workers at these companies (Schneider & Harknett, 2019a; Storer et al., 2020). The survey was fielded in English only. According to Facebook's statistics, the click- through-rate for the survey, or the percentage of those who clicked the Facebook Ad of those who saw the Facebook Advertisement, is about 3.2%. Of those who saw the Facebook Ad, about 1.3% filled out any information. These metrics are in line with previous research using Facebook Ads (Schneider & Harknett, 2019b).

After multiple imputation, the analytic sample includes 849 responses. For a more detailed discussion of the multiple imputation, as well as for a side-by-side comparison between data from this survey and other key sources, see Appendix D. However as noted in Schneider and Harknett (2019b), the demographics of the populations these samples are drawn from may be different from national samples of service workers, for instance because the CPS includes frontline workers in large and small firms. These issues are compounded in the current context, since the demographics of workers at the ten specific companies sampled in the current study may differ compared to a general sample of service sector workers. As a result, the analyses presented below are not weighted. In addition, Appendix D contains a comparison between the survey data and the data scraped from Glassdoor along key independent and dependent variables.

Dependent Variables

Job Satisfaction. Respondents are asked “All in all, how satisfied are you with your job at [Employer].”

Turnover Intentions. Respondents are asked “Taking everything into consideration, how likely is it you will make a genuine effort to find a new job within the next 3 months?”

Independent Variables

Agreement with How Company Treats Customers. Respondents are asked “How satisfied are you with how [Employer] treats CUSTOMERS?” This variable is meant to capture and further interrogate the significance of the organizational orientation for job quality ratings.

In order to size this effect, this paper also compares a series of other relevant theories of customer service work.

Emotional Labor. Respondents are asked “Do you agree or disagree with the following statement: In general, I find it difficult to put on a happy face for CUSTOMERS at [Employer].” This variable is meant to put the organizational question in perspective of another common view of customer service in sociology: a difficulty in conducting emotional labor.

Nice Customers. Respondents are asked “How often would you say your CUSTOMERS are rude to you?” This question is meant to capture the extent to which the perceived frequency of positive interactions with customers may improve job quality. This question and the following question capture the dyadic aspect of customer service interactions between frontline workers and their customers.

Rude Customers. Respondents are asked “How often would you say your CUSTOMERS are nice to you?” This question is meant to capture the extent to which the perceived frequency of negative interactions with customers may degrade job quality.

Controls

These analyses contain two sets of controls. First, fixed effects are included for occupation, company and state. These fixed effects play an important role in controlling for unobserved heterogeneity between employers and occupation which may be determining how service interactions are managed and what workers take away from these interactions. By controlling for these differences, I am able to isolate the localized process connecting a particular attitude about one's job to job satisfaction and turnover intentions.

Next, job quality controls are included for wage, the number of hours they usually work on the job, an indicator between zero and one indicating how variable the respondent's hours are over a month, if they have worked on the job for less than year, and whether they are full time. I also include a count of the experience of six negative schedule characteristics a worker experienced in the past month: (whether the respondent had worked on-call, had a shift canceled, been told to stay late, been sent home early, been told to come in early, or been told to come in late) a respondent may experience. Human capital controls are included for whether they received at least some college education, and whether they are enrolled in school. In addition, demographic controls are included for age, race/ethnicity, gender, cohabitation, and whether the respondent has children. Finally, a control is included for how frequently a respondent interacts with customers.

Analytic Strategy

I use the survey data to run linear regressions for two outcomes: job satisfaction and turnover intentions. Standard errors in each regression model are clustered by company, since reviews were drawn specifically from separate companies. The estimates presented below are derived from a final model where all independent variables are included in an analysis. For the results from each stage of the regression analysis, as well as additional sensitivity checks, see Appendix D.

Descriptive Statistics

The table below highlights the questions asked in the survey that are included in the regression analysis (Table 3).

While roughly 25% of workers report that they find it difficult to put on a happy face for customers, only about 12% of workers say they are dissatisfied with how the company treats customers. Additionally, a large proportion of respondents, roughly 30% say they rarely or never have customers act rudely towards them. On the other hand, over 75% of workers say customers are nice on a daily basis.

Results

The figures below depict the average marginal effects of the regression for each response category in the dependent variable. With respect to job satisfaction, the figure shows that there are significant returns to satisfaction with how the company treats customers (Figure 2).

Frontline workers who are somewhat satisfied with how the company treats customers are .4 points more satisfied with their jobs. Extreme satisfaction with customer treatment is correlated with a nearly .9 point increase in job satisfaction compared to the neutral group. The standard deviation of the dependent variable, job satisfaction, is about 1.35. So, a .9 point increase in job satisfaction is an increase of two thirds of a standard deviation in the dependent variable. Frontline workers who do not agree with how the company treats customers, on the other hand, average a nearly half point decrease in job satisfaction.

In order to put this effect in perspective an increase in the perceived frequency of positive or negative interactions with customers is not associated with any change in job satisfaction, and thus are not presented. Next, I contrast the effects above with the more established theory of emotional labor (Figure 3).

Frontline workers who state they strongly agree that it is difficult to put on a happy face for customers are roughly a half point less satisfied with their jobs, and those who say they agree are .28 points less satisfied. In terms of standard deviation, compared to the two thirds standard deviation decrease corresponding to extreme dissatisfaction with how the company treats customers, extreme difficulty putting on a happy face is correlated with a .37 point decrease in the standard deviation. While these effects are sizable, they are nearly half the size of the coefficients associated with satisfaction with how the company treats its customers.

Turning to plans to search for a new job, the effects are even more strongly pronounced (Figure 4).

Workers who are extremely dissatisfied with how the company treats customers report a .73 point higher likelihood to search for a new job in comparison to their neutral counterparts. On the other hand, frontline workers who are extremely satisfied along these lines report a .77 point lower likelihood to conduct a new job search. The standard deviation of the dependent variable is 1.55, so the swing from least to most satisfied with how the company treats customers is associated with about a one standard deviation decrease in plans to search for a new job. For the less extreme counterparts the differences are also statistically significant, with workers somewhat dissatisfied with how the company treats customers .36 points more likely to search for a new job, and somewhat satisfied workers .44 points less likely to conduct a search.

In contrast, as shown in the figure below, the only significant effect of difficulty conducting emotional labor on turnover intentions is among those who strongly disagree that it is difficult (Figure 5).

These workers are .38 points more likely to say they intend to stay at their current job. This effect is roughly half the size of the corresponding response among those satisfied with how the company treats customers.

Once again, a change in the perceived frequency of positive or negative interactions with customers is not associated with different levels of turnover intentions. This, in combination with the smaller effect sizes among those who report difficulty conducting emotional labor, once again puts the organizational aspect of customer service in stark relief. Workers are clearly impacted by their views of how the organization treats its customers. This view is present in written Glassdoor reviews, the association between these written reviews and Glassdoor ratings, as well as in this survey.