Abstract
Service work is seen by many to be less generative of working-class consciousness than non-interactive labor. This article interrogates that hypothesis using an original survey (N = 177) of New York State workers. Deploying intrinsic indicators for the intensity of service interaction and for working-class consciousness, the study finds that both the former and major demographic features fail to predict the latter while managerial status, workplace pain and discomfort, union membership, and job insecurity do. This supports an emergent view that services are broadly continuous with other forms of wage work and an older one that work itself is central to the production of class consciousness.
Introduction
The mid-twentieth century was rife with predictions of an imminently more peaceful society. Following the catastrophes of two world wars, depression and holocaust, which many attributed to intractable class conflict both within and between capitalist societies, the urge to seek ‘the end of ideology’ through the ‘embourgeoisment’ of factory workers or the ‘coming’ of a harmonious ‘post-industrial society’ was perhaps understandable (Bell, 1960, 1973; Fourastié, 1949; Goldthorpe et al., 1969). These claims were often based on an evolutionary analysis of the workforce, particularly in the growth of the service or tertiary sector vis-a-vis manufacturing and agriculture (Clark, 1940; Touraine, 1969).
While the macro-stabilizing predictions of this school have been largely falsified by subsequent history, an aspect of its reasoning continues to thrive in the more focused study of interactive service work. Gabriel (2005: 20) sees such jobs as embodying ‘a radical reconfiguration of workplace relations’ away from worker–manager hostility; Leidner, among the first to formulate the ‘service triangle’ concept (following Benson, 1986), advances a contingency or even postmodern interpretation of service workplace interests (Leidner, 1993, 1996, 1999); and Bolton (2005) proposes a four-part typology of ‘emotion management’ that paints service jobs as transcending the conflict and inequality of their non-service counterparts. In sum, the idea persists that interacting with customers, clients or patients as part of one’s job deflates or is somehow unproductive of conflict-oriented consciousness and action.
Our focus in this study is on working-class consciousness, which Mann defines as identity, opposition, totality, and ‘the conception of an alternative society’ (1973: 13; italics in original). Burawoy, echoing Marx, argues that this consciousness – or its absence – ‘is generated at the point of production rather than imported into the workplace from outside’ (1979: 135). But if service workers do not ‘produce’ in the same way as those in manufacturing, construction or transport, many believe they should be even less likely to hold self-conscious working-class attitudes. We find evidence to the contrary. Deploying multiple survey measures of job-based service interaction and working-class consciousness among an original sample of 177 working adults in Nassau County, New York, our results suggest that service interaction plays no role in producing or obscuring class consciousness. Neither, it seems, do salient demographic features such as age, race, gender or nativity. But four general job characteristics do, and together explain more than a third of the variation in respondents’ working-class consciousness: managerial status (negative), union membership (positive), pain and discomfort (positive), and job insecurity (positive). These findings support Burawoy’s argument for the centrality of work in the formation of consciousness and undermine a pervasive assumption that outside factors, particularly race, gender, and nativity, are more influential. Additionally, they support an emergent view among labor scholars that despite its distinctive features, interactive service work is broadly continuous with its non-interactive counterpart, subsumed as they both are within the capitalist structure of waged employment (Bélanger and Edwards, 2013; Brook, 2009; Warhurst et al., 2009).
In what follows, we first review the literature on class consciousness, the labor process, and the peculiarities of services to identify three competing hypotheses on their impact on workers’ class consciousness. Next, we detail our survey methodology before proceeding to a four-part consideration of the results: bivariate analysis, multivariate analysis using our primary service variable, three alternate service variables, and weighted data. Though important variations emerge, our findings are broadly consistent across service indicators and weighted and unweighted models. They suggest that service interaction and demographic background have little to no effect on workers’ class consciousness while general job features do.
Class consciousness in the service workplace
Class consciousness is among the most controversial concepts in all of social science. Before considering its relation to service interaction it is thus necessary define it. Disputes pertain to three basic questions: who can have class consciousness? what is it? and how does it develop? This study deals with working-class (rather than capitalist, petty bourgeois or peasant) consciousness and an historical outline of that subject might include all those employed in manufacturing, construction, and transport, plus their dependents (Cherlin, 2014: 8). This fits with Marx’s early depiction of ‘the proletariat’ as ‘a class of labourers … enslaved by the machine’ (Marx and Engels, 1988: 61–62) and Poulantzas’s (1975) more rigid interpretation that all non-goods-producing workers are ‘petty bourgeois.’ The latter view is in a distinct minority today, with most class theorists grouping routinized service workers together with their goods-producing counterparts as part of the working class, broadly conceived (Braverman, 1974; Ehrenreich and Ehrenreich, 1979; Resnick and Wolff, 2003; Wright, 1985, 1997; Zweig, 2000).
If service workers are thus potential candidates for working-class consciousness, what does it consist of? In some places, Marx describes it almost behaviorally as the formation of organizations that can articulate and advance working-class interests (1972 [1852]: 516). Lukács builds upon this to define class consciousness as ‘the appropriate and rational actions “imputed” to a particular typical position in the process of production’ (1971: 51). But the gulf between such conceptions and received notions of psychological consciousness is too wide for some to bridge: ‘collectivities do not “have” consciousness,’ argues Wright, ‘since they are not the kind of entities which have minds, which think’ (1997: 193). Instead he deploys an attitudinal construct derived from Mann (1973) and similar to those used in other quantitative studies (Goldthorpe et al., 1969; Leggett, 1968; Vallas, 1987a, 1987b; Zingraff and Schulman, 1984). Mann’s construct has four components: … class identity – the definition of oneself as working-class, as playing a distinctive role in common with other workers in the productive process … class opposition – the perception that the capitalist and his agents constitute an enduring opponent to oneself … class totality – the acceptance of the two previous elements as the defining characteristics of (a) one’s total social position and (b) the whole society in which one lives. Finally comes the conception of an alternative society, a goal toward which one moves through the struggle with the opponent. (1973: 13; italics in original)
Some reject Lukács’s imputational model for other reasons. Thompson argues that ‘class-consciousness is the way in which [production-related] experiences are handled in cultural terms’ (1963: 9–10) and his followers believe it should be neither ‘imputed’ via structural analysis nor ‘measured’ with questionnaires (Fantasia, 1988, 1995; Katznelson, 1986). Instead, they believe it should be studied qualitatively in local contexts, eschewing the conception of ‘class consciousness’ as an abstract variable that can be studied comparatively or transferred from one national (or regional) setting to another.
These perspectives are not mutually exclusive, however: one can stipulate what class interests are and what actions advance them; one can assess individuals’ attitudes to see where they stand in relation to these; and one can study workers in situ to determine the processes that engender intersubjectivity, of whatever form, among them. Differences seem to revolve more around researchers’ preferred methodology – historical-comparative, survey or ethnographic – than theoretical substance. Though originally developed in the UK, Mann’s schema is most appropriate for our purposes given its intended international applicability (it was used in his original 1973 study to compare developments in the UK, France, and Italy) and its clearly delineated dimensions that are readily operationalized for survey research. We do not, however, view it as intrinsically superior to its alternatives.
Third is the question of how working-class consciousness develops. This primarily concerns the alignment or disconnect between objective class position and one’s subjective awareness. It can thus be re-phrased as: which social forces encourage working-class people to identify and act as such vis-a-vis other classes or, simultaneously, to identify and act otherwise? Culture, ideology, politics, as well as race, gender, and nationality all conspire, but the workplace plays an especially important role. Burawoy (1979) contends it is the prime place for capitalist relations to be either revealed or concealed and his research in a Chicago-area machine shop uncovered the impact of both occupational divisions and managerial relations on worker consciousness. These same dimensions –work organization and employment relations – have since gained traction as the defining features of workplace regimes (Burawoy, 1985; see also Spenner, 1983; Vallas, 1987a). Organization involves the degree to which tasks are collectivized, the extent to which they are routinized, and how closely managers dictate and monitor their completion. Increases along these vectors may engender working-class consciousness with collectivization encouraging solidarity, control encouraging opposition, and routinization heightening workers’ boredom and irritability (Aronowitz, 1972; Edwards, 1979; Johansson, 1986). Employment relations such as wages, benefits, and job security also shape worker consciousness. Jacoby (1997) traces the rise of American ‘welfare capitalism’ in which employers provided generous wages and benefits in order to keep unions out; Kochan et al. (1994), Lichtenstein (2006, 2009), and Moreton (2009) detail the development of less generous but equally effective models aimed at mitigating opposition and unionization. Burawoy (1979), however, finds that unionization itself can sometimes transform relations into a fair ‘game’ that corrals worker militancy. In sum, both the organizational and employment-relational aspects of labor processes are likely to shape workers’ class consciousness in relatively independent ways.
But service work is somehow different, many argue. Fourastié (1949), Touraine (1969), and Bell (1973) believed macro-level transitions to ‘post-industrial’ employment would mitigate working-class consciousness or refract it into status-group consciousness. Mills conceptualized ‘white collar’ employees – including frontline service and clerical workers – as distinct from the working class and possessing a ‘marketing mentality’ hardly at odds with capital (1951: 182). Poulantzas (1975) also grouped all non-goods-producing employees in what he termed the ‘new petty bourgeoisie’ upon which he projected an individualist, non-conflictual, and politically reactionary consciousness. Hochschild, however, greatly refined the study of services per se with her concept of emotional labor: ‘labor requir[ing] one to induce or suppress feelings in order to sustain the outward countenance that produces the proper state of mind in others’ (1983: 7). She proposed that interpersonal services prioritize abilities that are only secondary in other occupations: initiating interaction with clients, reading their emotional displays, and adjusting one’s responses to complete a given task, be it selling shirts, serving meals or inserting intravenous tubes without undue anxiety. She differentiated between jobs that call for ‘surface acting’ – brief, superficial displays – and those that called for ‘deep acting’ – extended interactions that require feeling one’s displays to some degree. Lopez (2006) expands this differentiation to the institutional level, conceptualizing alienated ‘emotional labor’ as typical of for-profit environments, and more humane ‘organized emotional care’ as typical of non-profit or public ones.
The effects of emotional labor on workers’ class consciousness, however, have only been approached at the margins, with the sole exception of Jones (2001), who found a modest positive relationship between ‘caring labor’ jobs and working-class consciousness using Wright’s 1980 data. Hochschild and others (Erickson and Ritter, 2001; Erickson and Wharton, 1997; Hart and Warren, 2013; Wharton, 1993, 1999) consider the ‘negative consequences’ of service interaction which may include the ‘fusion of self and work role’ or the ‘estrangement’ of the two, causing ‘emotive dissonance,’ feelings of ‘inauthenticity’ and burnout. Besen-Cassino (2014), however, finds a more positive fusion among youth service workers in affluent communities who value the social aspects of their jobs and learn to perform ‘aesthetic labor’ that entwines their identities with the brands they sell (see also Hall and Van den Broek, 2011; Nickson et al., 2011; Warhurst et al., 2000). Misra and Walters (2016) identify similar processes among fashion retail workers, but find that branded identities can also be a source of resistance to managerial attempts at speed-up or simple incompetence; furthermore, they find that such subjective fusion fails to mask workers’ low pay, meager benefits, and job insecurity. The implications for working-class consciousness are somewhat ambiguous: insofar as service workers become alienated and dissatisfied, this may enhance their opposition to management and openness to change. A ‘fusion of self and work role’ might also enhance class identity if it creates a durable occupational mindset and community, but the findings of most empirical researchers seem to point in the opposite direction – either towards an uncritical bonding of workers with the goals of the brand of employers or towards a generalized burnout that leads them to individualize problems or dissociate from their jobs, ‘maintain[ing] a healthy distance between self and work role’ (Wharton, 1999: 164).
Emotional labor is also highly gendered. Hochschild emphasized its frequent ‘feminization’ in Western societies, 1 while England (1992) developed the related concept of ‘care work’ to describe how many interactive duties are transplanted into waged contexts from the domestic sphere and typically undervalued (see also Guy and Newman, 2004; Hartmann, 1981; Himmelweit, 1999). Though Cockburn (1983) and Federici (2004) argue that gender has always been fundamental to the capitalist division of labor, service work researchers emphasize its outsized role in structuring societal perceptions, demographics, and even modes of resistance of those working in this sector (Crocker and Clawson, 2012; Milkman et al., 1998). Insofar as emotional labor is sex-typed and devalued, an ensuing gender solidarity and consciousness may augment union or working-class consciousness (Cobble, 1991). But to the extent traditional gender roles are naturalized and dominate workers’ self-perceptions, as Moreton (2009) found among female Walmart workers in the 1970s, this may mask the reality of their exploitation and blunt the emergence of class consciousness.
Leidner (1993, 1996) and Benson (1986) argue that services also differ in their triadic structure. Whereas in other work contexts the recipient of the product is removed in time and space from its production, leaving workers and managers facing each other alone, service environments involve customers, patients or clients in these relationships. Leidner found that some fast-food workers allied with managers or managerial protocol to calm unruly customers, while door-to-door insurance salespeople frequently colluded with customers to bypass managerial protocols when unnatural or superfluous. Taylor and Bain (1999) detail how some service managers use customer reports to discipline workers, displaying the possibility of customer–manager alliances. This ‘Bermuda triangle’ (Bolton and Houlihan, 2010; Lopez, 2010) thus offers more possibilities for alliance and conflict and leads us to expect lower class consciousness among service workers since they are more often on the managers’ side or have recourse to customers as short-term allies.
Bolton (2005, 2009) and Leidner (1993, 1999) advance conceptions of service work as outside the realm of capitalist class relations. Leidner states that ‘which of the three parties [workers, managers or customers] to service interactions achieves their goals depends on the resources available to them in particular settings,’ appearing to place all three on a level field with outcomes determined contingently (1993: 174). Bolton develops a four-part typology of ‘emotion management’ which she distinguishes from Hochschild’s more grounded concept. Two of her types – the ‘pecuniary’ and the ‘prescriptive’ – are bound by workplace prerogatives, while the other two – ‘presentational’ and ‘philanthropic’ – appear as more general dimensions of human interaction that permeate or transgress the boundaries of employment. If accepted, her conception would again suggest lower class consciousness among service workers since their objective constraints are not determined by ‘class’ but by interpersonal networks and contingent negotiations.
Several recent theorists, however, take issue with this approach. Instead of a sharp break between service and non-service jobs, Brook (2009), Warhurst et al. (2009), Bélanger and Edwards (2013), and Ikeler (2016a, 2016b) see continuity insofar as both are subsumed under the employment relationship. Taylor and Bain (2005: 264) argue that ‘the customer–management–agent “triangle” is not equilateral’ but heavily biased toward managerial prerogatives. While still acknowledging the distinctive features of working on people rather than things, Bélanger and Edwards argue that ‘a deeper understanding of service work is possible only if the worker–customer interaction is conceived as part of the social structure that shapes it, namely the employment relationship’ (2013: 433). Though not addressing working-class consciousness directly, the views of these theorists lead us to expect no primary association – positive or negative – with service work. Their latent hypothesis mirrors the findings of Wharton with regard to burnout, which was ‘better predicted by more conventional job characteristics, such as autonomy and number of hours worked, than by emotional labor’ (1999: 165).
In sum, the large and growing literature on service work mainly suggests a negative effect on working-class consciousness due to the deferential nature of interaction, the triangle of parties, and the transcendent relations between workers and clients (echoing Bell’s hypothesis). A minority view suggests the opposite when service work is highly gendered and/or allows for meaningful relationships between workers and clients that the former seek to defend against commodification (Hartmann, 1981; Lopez, 2006). A third view disputes the idea that service and non-service jobs are discontinuous, suggesting instead that general job characteristics such as autonomy, wages, job security, unionization and the like play determining roles in the production of consciousness or other outcomes, rather than its interactive content. These views constitute three possible answers to our central research question: how does service interaction impact workers’ class consciousness? They can be summarized as the following three hypotheses:
H1: Increased service intensity leads to decreased working-class consciousness among job holders.
H2: Increased service intensity, particularly in highly sex- or ethno-typed occupations, leads to increased working-class consciousness among job holders.
H3: Service intensity, in and of itself, has little to no impact on job holders’ working-class consciousness.
Methodology
To interrogate these competing hypotheses we chose to field an original survey among a sample of working adults in the United States. We opted for this more labor-intensive method over secondary data analysis for two reasons: (1) because existing attitudinal surveys, such as the General Social Survey, do not measure most elements of working-class consciousness; and (2) because those that do, such as Wright’s (1985, 1997) which was used by Jones (2001), are dated and provide only rough occupational proxies for the service interaction variable. Given these limitations, we launched an effort to gather fresh survey data with full and accurate measures of service interaction and working-class consciousness.
Sample
Our sampling frame was restricted to Nassau County, New York since the authors’ institution is located there and it was thought more likely to generate higher response rates given residents’ familiarity. Nassau County’s workforce, which numbers about 660,000, has a higher median income (but also higher costs of living) than its national counterpart. It is overwhelmingly suburban and is situated within New York City’s metropolitan area, which has double the national unionization rate (Hirsch and Macpherson, 2017). In demographic and sectoral terms, Nassau displays equivalent shares of gender, age, and racial groups, as well as major industry employment shares, with the exceptions of the foreign-born population (twice as large as in the nation), manufacturing employment (half as large) and finance, insurance, and real estate employment (50% larger). While some of these factors – unionization, immigration, and location within a solidly Democratic state – may make Nassau County workers more ‘class conscious’ than the typical US worker, other factors – high incomes, less manufacturing and more finance employment, and the historical dominance of the Republican Party on Long Island – point in the opposite direction.
In two waves during fall 2015 and spring 2016 a 40-item questionnaire was mailed out to a randomized list of 1000 respondents identified by direct mail as employed adults residing in the county. A one-page cover letter was provided on university letterhead that detailed the goals, procedures, and anonymity of the study, as well as respondents’ rights; subjects were also given a small monetary incentive and stamped, pre-addressed return envelopes. Forty-five questionnaires were returned due to changed addresses or simply with the note ‘retired’ – these were subtracted from our denominator. In total, 177 respondents returned the surveys (although 14 were missing some data, most of which we imputed subsequently). This yielded a response rate of 18.5%. As Table 1 shows, our sample is older, whiter, more educated, more female, more native-born, and slightly higher earning – though not more unionized – than the Nassau County workforce. We control for these biases through multi-factor adjustment cell weighting, discussed in our results section. There we see that when our cases are weighted to approximate the parameters of the broader county workforce, regression results are still largely the same.
Demographics of study sample and Nassau County, NY workforce.
Figures from American Community Survey (ACS) 2010–2014 merged data.
Table 2 displays the occupational characteristics of our sample. Here we see that approximately 40% of respondents worked in jobs typically coded as ‘frontline’ or interactive service: personal care (2.3%), healthcare practitioner or support (11.3%), education, training or library (15.3%), community or social service (5.1%), and sales (6.8%). But this is only a rough measure of service intensity that likely hides considerable variability across these and other occupational categories. Therefore, although we later deploy this occupation-based variable as an alternate, our primary measure of service interaction is an intrinsic assessment of respondents’ job content.
Occupational characteristics of study sample.
Measurements
Table 3 displays mean scores (within our sample) for the two main variables under investigation: job-based service interaction and working-class consciousness. The first of these, our independent variable, was constructed as a simple summated index of two five-point Likert scale prompts: (1) ‘To what extent does your job require you to interact with members of the public, including customers, clients, students or patients?’ and (2) ‘To what extent does your job require you to form long-term, in-depth relationships with members of the public, including customers, clients, students or patients?’ Response categories for each were ‘never, rarely, sometimes, often, always’ coded as 0–4 and summated into a nine-point (0–8) index. The first of these was designed to directly assess the service intensity of respondents’ jobs without relying on secondary indicators such as their occupation (e.g., Jones, 2001). The second was designed to assess the depth or length of interactions between respondents and their recipients, if any, as outlined by Leidner (1993), Hartmann (1981), and Jones (2001) as an important dimension of interactive work. These two measures, which had a reliability coefficient (Cronbach’s alpha) of .74, were summated to create SvcInteract.
Descriptive statistics for workplace variables.
The problem with this index, however, is that it is highly skewed. Its skewness (–.95) and kurtosis (.06) deviate from acceptable limits so we transformed the index by squaring its values. This yielded a distribution closer to normality with skewness (–.29) and kurtosis (–1.32) within usable ranges. We refer to this transformed measure as SvcInteract2.
Our dependent variable, working-class consciousness, was constructed from seven different survey indicators designed to measure the four dimensions outlined by Mann (1973) and used by many since. They were:
ClassIdentity: ‘If you were asked to use one of four names for your social class, which would you say you belong in?’ (lower, working, middle, upper);
ClassAntagonism: ‘If you give them a chance, the people who run large corporations will try to take advantage of the workers’ (five-point Likert, strongly disagree to strongly agree);
ClassOpposition: ‘Some people feel the relationship between management and workers has something in common with team sports like football or baseball. If we make that comparison, which of the following views do you think is more realistic?’ (dichotomous: ‘workers and management on same team’ vs. ‘workers and management on opposing teams’);
WorkerControl: ‘Most decisions would be better if they were made by the workers themselves, not by supervisors’ (five-point Likert, strongly disagree to strongly agree);
JoinUnion: ‘If given the choice, I would join a labor union’ (five-point Likert, strongly disagree to strongly agree);
ClassPolitics: ‘All things considered, politicians help employers more than workers’ (five-point Likert, strongly disagree to strongly agree); and
SansProfitMotive: ‘It is possible for a modern society to run effectively without the profit motive’ (five-point Likert, strongly disagree to strongly agree).
Several of these prompts were borrowed directly from Vallas (1987a), who undertook a parallel study among communication workers. The first of these measures, while having four potential response categories, only had three within our sample (working, middle and upper class). To scale this up to the same five-point (0–4) value of the other measures, we made ‘working class’ = 4, ‘middle class’ = 2 and ‘upper class’ = 0. Similarly for the dichotomous variable ClassOpposition we made ‘workers and management on opposing teams’ = 4 and ‘workers and management on same team’ = 0. Once transformed in this way, the seven indicators had a reliability coefficient (Cronbach’s alpha) of .66 and were thus summated into a single index, ClassConscious, which displays a normal distribution.
Our questionnaire also included many additional indicators for control variables. These fall broadly into three categories: intrinsic aspects of respondents’ jobs, extrinsic aspects thereof, and demographic features. The intrinsic category includes two measures of task complexity, ComplexInfo (‘My job requires me to analyze and interpret information’) and ComplexVariety (‘On my job I get to do a number of different things’); two measures of autonomy, WorkAsTold (‘My job requires that I do things just the way I am told’) and WorkMeasured (‘The amount of work I do is carefully measured by the people above me’); one measure of work collectivization, WorkCooperate (‘My job requires that I cooperate with coworkers to get things done’) and one of working conditions, WorkPain (‘My job exposes me to pain, exhaustion or discomfort, including extremes of heat or cold’). All were measured using the five-point Likert coding, ‘never, rarely, sometimes, often, always.’ An additional indicator, Managerial, was composed as the sum of four dichotomous measures assessing respondents’ supervisory duties: whether they had the power to hire, fire, direct the activities or evaluate the performance of other employees. A score of 0 indicated they had none of these duties, a score of 4 indicated they had all.
The category of extrinsic job aspects includes eight measures: SelfEmployed (dichotomous with ‘work for wages or salary’), WorkplSize (number of employees), UnionMember (yes or no), JobTenure (years in current job), Hours/Week (number of hours), HourlyWage (dollars), FriendlyBoss (‘My immediate supervisor is considerate of the employee’s feelings’ – five-point Likert response) and JobInsecurity (‘How often do you worry you’ll be laid off and have to look for a new job outside the company entirely?’ – five-point Likert response). Descriptive statistics for these variables are also provided in Table 3.
Demographic measures included respondents’ Age, gender identity (Female), racial/ethnic identity (White, Black, Hispanic/Latino, Asian, or other), recoded as POC (Black and Hispanic), White and Asian dummy variables, nativity (USBorn), and Education (‘less than HS, HS or GED, some college/associate’s degree, bachelor’s degree, graduate or professional degree’), recoded as the dichotomous Bachelor’s. Descriptive statistics for these are provided in Table 1. In addition, we also asked respondents their industry of employment (‘Which of the following industries best describes the main product or service created in your workplace?’ with 12 response categories reflecting all two-digit NAICS industries) and occupation (‘Which of the following best describes what you do in your current job?’ with 14 response categories reflecting all two-digit SOC occupations). The second of these was used to create an alternate measure of job-based service interaction which we deploy further in our results section.
Results
Bivariate correlations
Our empirical analysis begins with a simple bivariate matrix of 24 key variables (Table 4). These include SvcInteract2, ClassConscious, our measures of intrinsic (labor process) and extrinsic (employment relational) job aspects, as well as demographic controls.
Bivariate correlation matrix.
Table 4 shows that ClassConscious and SvcInteract2 are not significantly related and display only a weak negative coefficient. Preliminarily, this seems to speak for the third (or null) hypothesis about the relation between service interaction and working-class consciousness. Yet proceeding down the second column, we see nine other variables significantly correlated with ClassConscious: UnionMember (positive), HourlyWage (negative), WorkAsTold (positive), WorkMeasured (positive), WorkPain (positive), FriendlyBoss (negative), JobInsecurity (positive), and Bachelor’s (negative); Age also approaches significance and shows a negative correlation. Each of these factors relate to ClassConscious in expected directions, e.g., positively for UnionMember and WorkAsTold, negatively for HourlyWage and FriendlyBoss. But conspicuously insignificant are JobTenure, WorkCooperate, WorkplSize, the two measures of complexity, and all but one of the demographic measures (age, gender, nativity, and race) which past studies have found meaningfully shape class consciousness.
The patterns found at this bivariate stage were used to construct parsimonious multivariate models while excluding from the same models control measures with a high likelihood of collinearity. For example, though both WorkAsTold and WorkMeasured were significantly correlated with ClassConscious, they were also strongly correlated with each other (Pearson’s r = .50), and since the former was also closely related to Managerial, which itself had a stronger connection with ClassConscious, we did not include WorkAsTold in our regression models. Similarly, Age and JobTenure were not used simultaneously, but rather alternated in sequential models, given their equally strong correlation (Pearson’s r = .50). Overall, a bivariate Pearson’s correlation of .30 or higher was used as our threshold for mutual exclusion at the multivariate stage.
Regressions with SvcInteract2
Though bivariate techniques showed no relation between SvcInteract2 and ClassConscious and thus appeared to support Hypothesis 3, this could easily be due to intervening factors which, if controlled for, might alternately yield support for Hypotheses 1 or 2. Multivariate analyses, of which we conduct three series, account for this possibility.
Table 5 displays our first series of regressions on ClassConscious using SvcInteract2 as the independent variable. Model 1 includes only this measure and explains virtually none of the variance in ClassConscious (Adj. R2 = –.01). Model 2 then adds three intrinsic measures that were found to be both significant and non-collinear at the bivariate stage: Managerial, WorkMeasured, and WorkPain. Altogether, this four-measure model explains 20% of the variance in ClassConscious and two of those three additional factors maintain significance in the expected directions – negative for Managerial and positive for WorkPain. Interestingly, WorkMeasured loses significance when included alongside these variables and never regains it in later models.
Linear regressions on ClassConscious using SvcInteract2.
Unstandardized coefficients; standard errors in parentheses.
p ≤ .10; *p ≤ .05; **p ≤ .01.
Model 3 substitutes five extrinsic job measures for those three intrinsic ones: UnionMember, JobTenure, HourlyWage, FriendlyBoss, and JobInsecurity. Combined with SvcInteract2, these five measures explain even more, i.e., 32%, of the variance in ClassConscious than Model 2 and all four controls that were significantly related to ClassConscious at the bivariate level maintain that status and their direction – positive for UnionMember and JobInsecurity, negative for HourlyWage and FriendlyBoss.
Model 4 simply combines Models 2 and 3 and achieves a modest increase in overall explanatory power (Adj. R2 = .37). Six of its nine predictors achieve significance and none changes direction vis-a-vis previous models or the bivariate matrix. JobTenure still fails to gain significance while WorkMeasured repeats its underwhelming performance. But most importantly, even in this powerful model, SvcInteract2 fails to play any meaningful role.
One could argue that this is due to the absence of demographic factors in Model 4. Yet Model 5, which adds five such measures (Age, Female, POC, USBorn, and Bachelor’s) while subtracting JobTenure (given the danger of collinearity with Age), deflates this. Model 5 is equally as predictive of ClassConscious (Adj. R2 = .37) as Model 4 and SvcInteract2 remains insignificant and ineffective.
Model 6 then examines the power of demographic controls independently of intrinsic or extrinsic job features. Though it does enable Bachelor’s to regain its bivariate significance and for Age to approach that status, it explains very little variance in ClassConscious (Adj. R2 = .03) and does not alter the irrelevance of SvcInteract2. In sum, this first series of regressions shows that while multiple intrinsic and extrinsic aspects of respondents’ jobs – managerial status, physical discomfort, union membership, wages, security, and relations with supervisors – explain class consciousness to a meaningful extent, the interactive dimension of their work, as well as their background characteristics, play almost no role. This provides additional support for Hypothesis 3 and against 1 and 2.
Regressions with alternate service variables
Of course, one could object that our rather skewed index of ‘service interaction’ does not accurately measure what it should. Because of this, the reasoning goes, it fails to display significance with ClassConscious in these models and there is yet hope for Hypotheses 1 or 2. Table 6 grapples with this possibility by re-running the two most powerful models from the previous series with three alternate service interaction measures. The first of these, InteractPublic, is a simplified version of the first component of the SvcInteract variable. Likert-scale responses to the question ‘To what extent does your job require you to interact with members of the public?’ were themselves highly left skewed, so we dichotomized the measure by recoding responses of ‘often’ and ‘always’ as 1, and all others (‘never,’ ‘rarely,’ ‘sometimes’) as 0. Using this measure in Models 1 and 2, we see marked similarity to Models 4 and 5 from Table 5 – namely, the service predictor remains non-predictive while the same group of significant controls maintain that status, their direction and approximate strengths; furthermore, the overall models are identical in explanatory strength (Adj. R2 of .37) to their counterparts in Table 5. The only noticeable difference is the raised correlation coefficients for InteractPublic vis-a-vis SvcInteract2, but this is likely an artifact of the former’s simplified coding as shown by the similar Beta weights of each variable in their respective models (.09 for SvcInteract2 and .07 for InteractPublic).
Linear regressions on ClassConscious using alternate service variables.
Unstandardized coefficients; standard errors in parentheses.
p ≤ .10; *p ≤ .05; **p ≤ .01.
Models 3 and 4 deploy the other half of the original SvcInteract as independent predictor – respondents’ answers to the question, ‘To what extent does your job require you to form long-term, in-depth relationships with members of the public?’ The distribution of responses to this (‘never, rarely, sometimes, often, always’) was broadly normal and did not require transformation. But again, use of this alternate measure failed to yield different results: the explanatory power of Models 3 and 4 are almost identical to Models 1 and 2 (in this sequence) and 4 and 5 (in the previous one); all the old significant predictors – Managerial, WorkPain, UnionMember, HourlyWage, and JobInsecurity – remain largely unchanged; and InteractLongTerm fails to achieve significance while exhibiting a lower Beta weight (.05) than either of its previous alternatives.
Models 5 and 6 deploy an entirely different measure of service work composed from our survey question about respondents’ occupation: ‘Which of the following best describes what you do in your current job?’ Six of the 14 answer choices we designated ‘service intensive’ – personal care, food preparation or serving, healthcare practitioner or support, education, training or library, community or social service, and sales – and assigned their respondents a value of 1; all others received 0. This dichotomous measure, SvcOccupation, was strongly, positively and significantly correlated with SvcInteract2 (Pearson’s r = .39**), InteractPublic (.52**), and InteractLongTerm (.28**), yet derived from an entirely different survey indicator. When included in regression modeling, however, it yielded nothing new: adjusted R2 of .37 and .36, respectively, previously significant controls still significant in the same directions and with analogous strengths of correlation, and SvcOccupation itself not significant. In fact, its Beta weight dropped even further – to .01 – compared with those of its alternates.
These results buttress those seen in Table 5: service interaction, variously defined, is not significantly related to working-class consciousness, nor are workers’ demographic backgrounds, while six intrinsic and extrinsic job features are. Hypothesis 3 remains standing while Hypotheses 1 and 2 come in for further doubt.
Weighted regressions
Another possible objection concerns the likelihood of sampling bias. Our small sample diverges from the Nassau County workforce with regard to age, race, gender, nativity, wages, and education, as seen in Table 1. To control for this potential non-response error, we used adjustment cell weighting to create case weights for each of our 177 respondents using those six variables as measured against the parameters of 2010–2014 merged ACS data on the Nassau County workforce (accessed through the Integrated Public Use Microdata Series – Ruggles et al., 2015). Applying these weights to our dataset, we then re-ran the two most powerful regression models, originally identified in Table 5, for each of our four service interaction variables. This generated the eight regression models seen in Table 7.
Weighted linear regressions on ClassConscious.
Unstandardized coefficients; standard errors in parentheses.
p ≤ .10; *p ≤ .05; **p ≤ .01.
The first thing noticeable across all weighted models is that, again, none of the service interaction terms achieve significance. Second, the explanatory power of all models is largely the same as before (Adj. R2 between .36 and .38). The only caveat is that the relationship between models with and without demographic controls is reversed: whereas previously those without such controls were slightly more powerful (e.g., Adj. R2 of .36 vs. .35), once demographic weights are applied they are slightly less powerful (e.g., Adj. R2 of .37 vs. .38). Most previously significant controls (Managerial, WorkPain, UnionMember, and JobInsecurity) retain their significance and directions of correlation, but HourlyWage, which was itself a component of the weighting procedure, and FriendlyBoss, which was not, lose significance across all weighted models. This suggests that the unweighted significance of the latter two may have been due to sample bias, but it simultaneously suggests that the significance of the first four – as well as the non-significance of SvcInteract2 and its alternates – was not.
In sum, although our use of adjustment cell weighting does not completely obviate the objection of sampling bias, it provides a substantial piece of evidence against it, further supporting the significance of our unweighted findings – namely, that service jobs have no special influence on their performers’ class consciousness beyond the more general features of workplace regimes. Hypothesis 3 is thus consistently supported across all multivariate analyses.
Conclusion
This study set out to interrogate the connection between service interaction and working-class consciousness. Though not always directly stated, the leading interpretation of services as markedly different from other job sectors and, in many cases, as removed from or transcendent of the latter’s antagonisms, strongly implies a negative hypothesis: all else being equal, interactive service workers should be less class conscious than their non-interactive peers. Our findings contradict this expectation (Hypothesis 1), as well as its less frequent positive converse (Hypothesis 2). In every single model, using weighted and unweighted data and four different independent measures, service interaction was found to be insignificant with respect to working-class consciousness – it did not predict whether workers were predisposed to join together to advance their interests vis-a-vis employers, capitalists or the state. In a broader sense, it failed to predict how they even perceived their interests. This constitutes robust support for Hypothesis 3.
What was perhaps more surprising was the simultaneous failure of key demographic variables – race, gender, and nativity – to predict working-class consciousness. Intersectional theory, which emphasizes the importance of these features in explaining social outcomes (Collins, 1990; Crenshaw, 1989; Harnois, 2010), has become increasingly influential and advances a viewpoint we are otherwise inclined to accept. Why its main tenet is not supported here is difficult to interpret, but our results point to one promising explanation: work in general, not its service content, may be a homogenizing force among workers of different backgrounds and the primary driver of class consciousness. It is thus central to the production of working-class consciousness. This perspective is the essence of Burawoy’s (1979) proposal that the labor process shapes class consciousness more definitively than workers’ out-of-work identities. It also dovetails with Silver, who, while arguing for ‘the centrality of race, ethnicity, gender, and nationality to class formation’ finds a pattern of ‘strikingly similar labor movements emerg[ing] among mass production autoworkers in vastly different cultural and political settings’ due to what she considers the standardizing impact of first Fordist, then just-in-time production processes (2003: 21, 32; italics in original).
Despite the insignificance of service interaction or demographics, our regression models are quite powerful: the full versions explain more than a third of the variation in working-class consciousness and the bulk of this can be attributed to four key variables – managerial status (which maps closely with worker autonomy), workplace pain and discomfort, job insecurity, and union membership, which, in the US, is often an objective characteristic of workplaces rather than an expression of workers’ subjective choice. All of these are work-related, with two pertaining to intrinsic job features (Managerial and WorkPain) and two to extrinsic job features (UnionMember and JobInsecurity). For all but Managerial, the relationship was positive: more pain and discomfort, being a union member, and less job security predicted greater working-class consciousness while increased managerial duties predicted less.
This cuts across claims that interactive service work is uniquely different from or ‘transcends’ the conditions of non-service work, as claimed by Leidner (1993), Bolton (2005) and heavily implied by Bell (1973). While service jobs may place more emphasis on practitioners’ emotional capacities and less on their physical or purely intellectual ones, it likely draws on all three, as do assembly-line production and chemical engineering, if in different proportions. Service work thus appears to be continuous with other forms of wage labor, as argued by Brook (2009), Warhurst et al. (2009) and others, rather than sharply distinct. Our findings reinforce those of Palm (2017) that class consciousness is not inexorably on the wane and of Bélanger and Edwards that ‘not the immediate nature of [one’s] job, but rather the economic and social relations in which it is located … is critical’ (2013: 447).
Our findings differ from those of Jones (2001: 294), who plumbed a closely related question and found a positive correlation between ‘caring labor’ jobs (particularly ‘high-intensity’ ones) and class consciousness. Overall, his models were weaker, peaking at an R2 of .10, whereas ours peaked at an adjusted R2 of .37 in the unweighted and .38 in the weighted series. Furthermore, his measurement of ‘caring labor’ was entirely occupational – i.e., not based on respondents’ self-reported job content, as our measure of service interaction was – and came from a dataset (Wright, 1985) that is now almost 40 years old. In these ways, our study is both more rigorous with respect to measurement and more current with respect to data, despite its modest size. While our weighted results provide for inference to the Nassau County workforce, this is still a small universe within a much larger national and larger still international workforce. Because of this, our results can only be taken as suggestive.
The most promising pathways for future research would include widening the scope of data collection and expanding or deepening the measurement of service interaction, to see whether the same non-relationship with working-class consciousness still holds. A more derived pathway would interrogate the links between workplace experience, worker subjectivity, and collective action. Working-class consciousness as an ideational construct is one thing, as Fantasia (1988, 1995) reminds us, and which we found is predicted by work but not by its service content; joining a union, going on strike or protesting in public are riskier endeavors, and are often constrained by structural forces and political dynamics that may cut across – or may yet build upon – the attitudinal patterns found here.
Footnotes
Acknowledgements
We would like to thank David Monaghan for his assistance in the adjustment-cell weighting process, Ruth Milkman and Luke Elliott-Negri for their feedback on earlier drafts, and the anonymous reviewers for their constructive input.
Funding
This research was supported by a Professional Development Grant of $1830 from the United University Professionals (SUNY faculty union).
