Advancing human resource management scholarship through multilevel modeling

Abstract

HRM systems are an organization-level construct that affect outcomes at the firm, unit, and individual levels of analysis. The multilevel nature of the field creates a need for both theoretical and empirical modeling that cuts across levels to effectively understand the linkages between HRM systems and various operational and financial performance outcomes. Ordinary least squares (OLS) regression which is designed to analyze the same level of data is not suited for analyzing such hierarchal data. Multilevel modeling accounts for variance among variables at different levels; dealing with sources of errors more rigorously than OLS. Multilevel structural equation modeling separately estimates between and within effects, takes into account measurement errors and allows for criterion variables that are situated at higher levels. Thus, multilevel modeling significantly advances HRM research by more accurately predicting HRM effects and estimating complex HRM models. The articles included in this collection demonstrate the value and application of multilevel modeling, both theoretically and empirically, to HRM research.

Keywords:

• Cross-level research
• hierarchical linear modeling
• human resource management
• multilevel modeling
• special issue

HRM is inherently a multilevel field of study. Complicated processes in the environment interact with organizational systems to affect outcomes at the firm, unit, and individual levels. HRM systems operate within complex organizational systems and structures that are further affected by macroeconomic forces and competitive forces that drive and alter implementation plans. These effects move through the layers of the organization to affect the performance of the firm. Simultaneously, these factors combine to affect micro-phenomenon like individual job performance, turnover, job satisfaction, organizational commitment, creativity, and citizenship behaviors. In sum, processes and systems in HRM cut across strategic and operational levels to affect both macro- and micro-oriented outcomes. The multilevel nature of the field is the impetus for this special issue, where we explore the importance of both building theory across levels and appropriately testing multilevel models. In this special issue, we have worked to select papers that serve as strong examples of both multilevel theory building, as well as, multilevel empirical modeling.

Multilevel modeling was first used in education and marketing research (Mathieu & Chen, 2011). HRM research has lagged behind in the application of this statistical technique as the majority of HRM research has historically been conducted at the single level of analysis (Shen, 2016). One of the reasons for this is a lack of theorization of the hierarchal nature of HRM practices (Ostroff & Bowen, 2000). As noted in Wright and Bosell (2002), the early literature segregated HRM research into macro- (strategic) and micro (functional)-areas and did not integrate the two levels of analyses effectively.

Recent trends have witnessed a growing interest in conducting HRM research at the cross-level of analysis (i.e. Jiang, Takeuchi, & Lepak, 2013; Ostroff & Bowen, 2000; Snape & Redman, 2010; Wright & Bosell, 2002). This collection reflects this trend. Multilevel modeling advances HRM research in at least two important ways. First, research conducted at the same level of analysis does not account for the salient contextual effects of HRM practices nested within organizations. Traditionally, hierarchal HRM data were analyzed using either disaggregation or aggregation approaches. The disaggregation approach assigns the mean score across units to individuals while the aggregation approach lifts the lower level variables to the higher level. In short, the disaggregation approach ignores between-group variance. In contrast, the aggregation approach neglects individual differences within units. Therefore, neither approach is well-suited to multilevel models. Multilevel modeling addresses these limitations (Osborne, 2000). Multilevel modeling estimates within and between effects simultaneously, therefore helping predict the HRM–employee or organizational outcome relationships more accurately.

Second, the availability of multilevel structural equation modeling (MSEM) techniques enables researchers to test complicated HRM models so as to better understand the complexity in the social and psychological processes involved. MSEM helps explore the effects of lower level predictors, such as employees’ perceptions of HRM systems, individual work attitudes and behaviors, psychological climates, and demographic factors, on the outcome variables situated at higher levels, such as employee collective attitude and behavioral outcomes, organizational performance, organizational climates, etc. A typical example is that HRM influences organizational performance through its effect on individual employee work behavior, the so-called 2-1-2 model (Preacher, Zyphur, & Zhang, 2010; Zheng, Zyphur, & Preacher, 2009).

Multilevel theory development and empirical modeling represent important advances for the field. Such approaches move the field forward by helping to build more robust models that account for common misspecifications in traditional methods, building avenues for more advanced theory building, and providing a means to test more complicated theories grounded in both psychological and sociological traditions. We further discuss these advantages below, before previewing the articles contained in this special issue.

Multilevel theory development

Many areas of social science research involve models of parameters that vary at more than one level, (i.e. there are micro-units nested in macro-units). For example, in education, students are nested in schools. Student academic results, therefore, are determined by individual study efforts and the school teaching quality shared by all students. In marketing, consumer behaviors can be influenced by individual factors and brands of products. In health sciences, people’s health can be affected by personal habits and common geographical factors (Rice & Leyland, 1996). HRM research is not an exception: the attitudes, behaviors, and performance of organizational members may be influenced by both personal factors and HRM policies and practices adopted by organizations (Ostroff & Bowen, 2000). Although individual perceptions of HRM implementation tend to be different within the same organization, the effects of HRM policies on employees (i.e. attitudinal and behavioral consequences of HRM) are inherently dependent and organization specific.

An argument can be made that the unique contribution of the organizational sciences in general is in explaining the dynamics of disparate phenomena at multiple levels (Behling, 1978; Rousseau, 1985). To meet this promise, organizational scholars must develop theories that cross-levels of analysis to explain complex social and psychological phenomenon. This is certainly the case in HRM research, where an increased emphasis on strategic HRM (i.e. Bowen & Ostroff, 2004; Guest, 2011) has led to a heightened need to understand the complex nature of the multilevel effects of HRM systems. While the empirical side of multilevel modeling has received some attention, the tools and technologies need to be matched by rigorous cross-level theorizing.

Multilevel theoretical models should establish the validity of the construct under investigation at the appropriate level of analysis (Arthur & Boyles, 2007). This infers the need to properly define constructs at the level of policy and practice, as well as develops a need to better understand the joint effects of HRM systems. The fact that many systems are implemented in different forms at the unit level creates the need to better understand implementation conditions (Nishii & Wright, 2008). In addition, proper theoretical development of multilevel models requires that theoretical connections be drawn from the organizational level to the individual level and vice versa. Such theorizes involves carefully specifying the means through which organizational practices affect individual attitudes and behaviors, which then in turn affect performance outcomes. For HRM research to continue to build useful knowledge these theoretical linkages need to be clear and well-grounded in existing understanding of human and organizational behavior. In this special issue, we present a number of studies that provide rigorous theory development across levels of analysis.

Methodological advantages of multilevel modeling

The inherent multilevel nature of the field means that theory-building, research design and analytic techniques need to be leveraged to properly understand the complex phenomena. This also creates a dependence on hierarchal data, which means that modeling needs to take into account the nature of the data in order to avoid deflated standard errors and inflated values of model fit or correlations (Raudenbush & Bryk, 2002). Regular OLS regression is designed to analyze the same level of data and is therefore inappropriate for analyzing multilevel data. Reliance on OLS leads to gross errors of prediction, hence is unfit for analyzing nested data (Rowe & Hill, 1998; Snijders & Bosker, 2012). Multilevel modeling incorporates both lower level and higher level predictors, and separates and estimates both within-unit and between-unit relationships so as to differentiate variance among variables at different levels (Osborne, 2000). As such, multilevel models handle sources of errors more rigorously than OLS (Rowe & Hill, 1998). Theoretically, multilevel models may involve parameters at various levels, but the most common multilevel models have two levels, as shown in all the studies in this collection. Multilevel models have a number of names including hierarchical linear models, nested models, mixed models, random coefficient models, random-effects models, random parameter models, split-plot designs, and variance component models. Despite these different names they all, in essence, deal with nested or hierarchical data (Raudenbush & Bryk, 2002). Moreover, multilevel regression can also be used to analyze repeated measurements as measurement occasions can be conceptualized as nested within individuals (Luke, 2004).

Traditional multilevel modeling data analysis is performed by testing a range of hierarchical models, namely the null model, the random intercepts model, the intercepts-as-outcome model, and the slopes-as-outcomes model (Snijders & Bosker, 2012). A null model estimates whether Level-2 predictors influence the residual variance in the individual-level model. The results determine whether multilevel modeling is appropriate for the data-set. A random intercepts model estimates the predicting power of Level-1 predictors for the Level-1 criterion variable. An intercepts-as-outcomes model is run to examine the cross-level effect of the higher level predictor on the Level-1 outcome variable. Finally, a slopes-as-outcomes model estimates cross-level interactions of Level-1 and Level-2 predictors on the Level-1 outcome variable. These procedures are adequate for testing a simple multilevel model without cross-level mediation, but not for testing meso-mediation.

Traditional multilevel modeling meso-mediation is tested by comparing the cross-level effects when the mediator is being controlled and those when the mediator is not being controlled (Freedman & Schatzkin, 1992). This procedure becomes unnecessary as the direct association between the independent variable and the dependent variable is not crucial to mediation (Kenny, Kashy, & Bolger, 1998). Also, this procedure does not take into account measurement errors (Preacher et al., 2010). Furthermore, traditional multilevel modeling is not able to test models with dependent variables situated at higher levels (Preacher et al., 2010; Zheng et al., 2009). Hence, it is suggested that MSEM should be used for testing meso-mediation relationships (Preacher et al., 2010; Zheng et al., 2009). Nevertheless, multilevel modeling is not without limitation. In particular, the requirement for sufficient sizes of both individual and group sample, i.e. over 30 units each with over 30 individuals (Maas & Hox, 2006; Raudenbush & Liu, 2000), is a constraint to some HRM researchers. While many computer software programs have the function for testing multilevel models, HLM is designed specifically for such a purpose. This advancement has refocused the nature of theory development and empirical testing to carefully consider the multilevel nature of HRM scholarship.

Contributions in this collection

The studies in this collection make both theoretical and methodological contributions. We begin with the study of Chidiebere Ogbonnaya and Danat Valizade, which tests a 2-1-2 multilevel mediation model in which high performance work practices (HPWPs) (level 2) affects organizational performance (level 2) through the mediation of employee outcomes (level 1). This model, also called a ‘bathtub model’ due to its steep vertical sides and relatively flat bottom, entails two kinds of multilevel effects: a 2-1 portion (the effect of a Level-2 predictor on a Level-1 mediator) and a 1-2 portion (the effect of a Level-1 mediator on a Level-2 outcome). HPWPs are distinct but complementary HRM practices aimed at developing employees’ skills for an effective organization. They typically include factors such as team work, involvement in decision-making, staff training, job discretion, and practices that optimize employees’ discretionary effort. There is growing evidence to suggest HPWPs promote organizational performance through positive employee outcomes; a phenomenon that shows researchers are now getting inside the ‘black box’ of HRM (Jiang, Lepak, Hu, & Baer, 2012). HPWPs shape positive employee attitudes and behaviors through empowering employees to commit their best effort toward promoting organizational performance. This assumption illustrates what has become known as the mutual gains perspective of HPWPs (Guest, 1997; Van De Voorde, Paauwe, & Van Veldhoven, 2012), the idea that HPWPs create a ‘win-win’ situation for both the organization and employees. In the mutual gains perspective, positive employee outcomes are seen as mediators in the causal chain between HPWPs and organizational performance.

The vast majority of past HPWP studies have, however, tested single-level mediation models; thus, ignoring the interdependences among employees nested within the same organization. Using secondary data from the British National Health Service, the study examined simultaneously both within and between effects. The study found the evidence for a direct positive relationship between HPWPs and employee job satisfaction and employee engagement. Moreover, both job satisfaction and employee engagement mediated a negative relationship between HPWPs and staff absenteeism, but the positive relationship between HPWP and patient satisfaction was mediated by job satisfaction only.

The next study by Shung Jae Shin, Inseong Jeong, and Johngseok Bae examined the roles of high involvement HRM (HIHRM) practices, intrinsic job motivation, and learning orientation in boosting production-line worker creativity. Using a sample of 3316 production-line workers from 240 manufacturing companies in South Korea, the authors theorized and tested a multilevel mediated moderation model in which organization-level HIHRM influences individual-level employee outcomes. The study suggests HIHRM practices were positively associated with intrinsic job motivation, which in turn resulted in employee worker creativity. Moreover, employee learning orientation moderated the relationships of HIHRM practices with intrinsic job motivation and worker creativity. The positive relationships existed only when employees had high levels of learning orientation. This study investigated cross-level relationships while avoiding the cross-level fallacy by following multilevel research methodology rather than aggregating individual-level variables or assuming the existence of HR competencies at organizational level (Ployhart & Moliterno, 2011). This study is one of few that explores the mediating mechanism through which HRM practices are related to worker creativity and the conditions under which this relationship is strengthened or mitigated. This finding supports the notion that employees may not respond to HRM practices in the same manner. Accordingly, consistent with the interactionist approach (Shalley, Zhou, & Oldham, 2004; Woodman, Sawyer, & Griffin, 1993), this study suggests that when studying the influence of HRM practices on employee behavior or performance, one should consider individual differences in dispositional traits to develop more fine-grained theories and understandings.

The study of Silvia Dello Russo, Daniele Mascia, and Federica Morandi investigated the interplay between perceptions of individual employees regarding HR practices and the variability of such perceptions within the department (i.e. HRM strength) and their effects. Specifically, the study addressed the following two research questions: What happens in contexts where HR practices are implemented with a high degree of variability? Should HR practitioners be concerned if employees differ considerably from one another in their perceptions of HR practices? HRM strength has often been conceived either as a pre-requisite for aggregating perceptions of HR practices at a higher level (Kehoe & Wright, 2013), or as a reinforcement of the effect of average HR practices on collective, rather than individual reactions (Katou, Budhwar, & Patel, 2014). However, individuals, each with their own different experiences, are all embedded in a common context that provides cues for sense-making, to the extent that this context can amplify or overshadow each individual’s own perceptions of HR practices.

The authors argue that HRM strength reduces reliance on idiosyncratic perceptions of HR practices and facilitates the emergence of collective responses. The study tested the hypotheses on a sample of 2821 health care professionals (i.e. nurses, head nurses, technicians, obstetricians, and allied health staff) nested in 44 departments of 27 hospitals. Cross-level analyses revealed that individual perceptions of HR practices positively predicted individual psychological proactivity climate, while they were moderated by HRM strength in the corresponding department. Moreover, idiosyncratic perceptions of HR practices predicted psychological proactivity climate when HRM strength is weak because ambiguous situations are interpreted by individuals based on direct experience; on the other hand, strong situations reduce the reliance on individual experiences making psychological proactivity climate more homogeneous with one another. This enabled the emergence of a collective climate for proactivity (i.e. individual perceptions of proactivity aggregated at the department level) to positively predict appropriateness of care. Appropriateness is a dimension of quality of care because when services are provided in the ‘right’ setting, they are offered in a technically correct way. The strengths of this paper consist in the multilevel nature of the data concerning HR practices, HRM strength, and proactivity climate, and the match with secondary objective data on department appropriateness of care provided by the Ministry of Health.

The paper of Viktoria Oppenauer and Karina Van de Voorde integrated macro-concepts, such as high involvement work system (HIWS), and more micro-related psychological processes to explore how HIWS impact individual-level employee health outcomes. More specifically, drawing on the job hindrances–challenges theory, the study examined the cross-level effects of enacted HIWS by line management on employee emotional exhaustion through two individual-level processes: (1) the capacity of HIWS to make work more intense and (2) the capacity of HIWS to make work more challenging. This study hypothesized that work overload and job responsibility mediate the relationship between HIWS practices (ability, motivation, opportunity, and work design HIWS practices) and employee emotional exhaustion.

The data for the study were collected from 360 employees and their line managers nested within 49 work units. Employees rated their feelings of work overload, job responsibility, and emotional exhaustion. The line managers from these work units rated the enacted HIWS practices. Results indicated that HIWS was positively related to work overload and job responsibility. In turn, job responsibility reduced emotional exhaustion, whereas work overload had a positive effect on emotional exhaustion. In addition, differential effects of HIWS practices on emotional exhaustion were found. Ability and motivation HIWS practices were positively related to work overload, and ability, motivation and work design HIWS practices were positively related to job responsibility. Opportunity practices were neither related to job responsibility nor to work load. Hence, the findings confirmed the idea that ability, motivation, opportunity, and work design practices affect employee outcomes in heterogeneous ways.

This study supports cross-level effects of HIWS on emotional exhaustion through two individual-level mediating mechanisms of job hindrances and job challenges, demonstrating that high involvement practices can be a double-edged sword. Therefore, when enacting HIWS practices within a work unit, they should be operationalized to empower employees to take on more responsibility on the one hand, but preferably without greater pressure and work load on the other hand, or at least combined with resources to mitigate the negative effects of work overload. The findings underline the importance of blending the occupational health psychology literature with the SHRM literature to develop a more refined model of the processes by which HIWS affects employee emotional exhaustion.

The paper by Luo and his colleagues explored the social and psychological processes which chief executive officers’ (CEOs) ambidextrous leadership, referring to an integration of both transformational and transactional leadership styles (Rosing, Frese, & Bausch, 2011), influences top management team (TMT) member ambidextrous behavior. It is argued that organizations capable of pursuing exploration and exploitation simultaneously have been suggested to obtain superior performance (CITES). Ambidextrous organizations, however, would have multiple and often conflicting goals, which pose considerable challenges for top managers. Past studies have largely neglected the role of individual managerial behaviors in cultivating an organization’s ability to explore and exploit simultaneously. Although ambidextrous leadership has received growing research attention, its organizational consequences have not been adequately explored (Zacher & Wilden, 2014). The study proposes that CEO ambidextrous leadership affects TMT member ambidextrous behavior via the social process of TMT behavioral integration, defined as the process through which ‘the group engages in mutual and collective interaction’ (Hambrick, 1994, p. 188). Second, the authors propose that TMT risk propensity will affect the extent to which ambidextrous leadership influences TMT behavioral integration, which in turn results in TMT member ambidextrous leadership behavior. The data for this study were collected from 214 TMT members and 59 CEOs. By testing a multilevel mediated moderation model, this study revealed that CEO ambidextrous leadership predicted TMT member ambidextrous behavior. Their relationship was mediated by TMT behavioral integration (i.e. collaborative behavior, information exchange, and joint decision-making). Moreover, the indirect relationship was moderated by TMT member risk propensity.

In addition to the studies examining the influence of HRM systems or practices on different levels of outcomes, this special issue includes three studies examining the factors of the adoption of HRM practices in organizations. Makhecha, Srinivasan, Prabhu, and Mukherji examined the multilevel gaps between intended, actual, and experienced HR practices in a hypermarket chain in India. Recent research has found that employees may have different perceptions of HRM practices from those of managers (e.g. Jensen, Patel, & Messersmith, 2013). More specifically, Nishii and Wright (2008) proposed a theoretical model of the intended, actual, and perceived HRM practices and emphasized that HRM practices perceived by employees may differ from actual HRM practices implemented in organizations which may differ from intended HRM practices designed by organizations’ decision-makers. Nishii and Wright have also proposed several factors that may explain the gaps between intended, actual, and perceived HRM practices, but very little empirical effort has been devoted to examine those factors. Makhecha and colleagues contribute to this stream of research by conducting a case analysis in seven retail units of a hypermarket chain in India. Based on the data from 41 interviews with managers and 128 interviews with front-end employees, they found that the ‘intended-actual-experienced’ gaps arise from implementer’s adaptation of HRM practices due to different understandings of the intent of those practices, the importance given to their contents, and the processes adopted in their implementation. This study provides a unique explanation for the multilevel gaps in HRM practices in organizations and highlights the need for further research on HRM implementation with particular focus on those gaps.

Most of multilevel research in HRM has examined the influence of HRM practices at the unit level on the individual level outcomes. However, firms are also nested within broader external context including strategic groups, industries, and countries. Therefore, it is theoretically and empirically important to examine how the contextual factors at the level higher than the firm level influence the adoption of HRM practices in organizations. Both Huang and Verma’s study and van Hoorn’s study contribute to this topic. Huang and Verma examined the industry- and firm-level determinants of employment relationships in China. They drew upon the institutional theory and the theory of HR flexibility to propose that employers’ choice of labor contract is influenced by industrial factors (e.g. competitive pressure and capital intensity) and firm characteristics (e.g. firm size, unionization, and ownership). Using a sample of 313 firms from 17 different industries within the manufacturing sector in China, Huang and Verma found that firms operating in industries with high international competitive pressure and low capital intensity are more likely to sign short-term labor contracts. Meanwhile, they also found that small firms, non-unionized firms, and non-state-owned firms tend to use short-term labor contracts. Their study is among the first few studies to focus on industry-level predictors of labor contracts and use a multilevel methodology to examine the cross-level influence. Their unique research design also allowed them to compare the relative influence of industry-level and firm-level factors on the choice of labor contract in firms.

Similar to Huang and Verma’s research, Van Hoorn focused on the use of job autonomy, which is a key workplace practices in organizations, and examined how the variation in job autonomy is accounted for by inter-industry, inter-country, and inter-supranational politico-institutional cluster variations. The author used the European Social Survey data collected from 119,932 individual employees from 62 industries and 30 countries and performed a unique four-level variance component analysis with job autonomy as the dependent variable at the lowest level (level 1, individual level) and politico-institutional clusters at the first highest level (level 4), countries at the second highest level (level 3), and industries at the third highest level (level 2) as predictors. His results suggest that industry difference explained much more variation in job autonomy than inter-cluster and inter-country dissimilarities even though the use of job autonomy did vary between country clusters and countries. This study not only contributes to the understanding of variation of a specific HRM practice but also sets a good example to conduct a variance component analysis that involves multiple levels.

Concluding remarks

The multilevel nature of research in the field of HRM demands a significant emphasis on understanding cross-level linkages between key predictors and outcomes. To advance the field greater attention is necessary to both the theoretical and empirical realities of multilevel phenomena. This special issue attempts to advance this conversation by highlighting both important theoretical and empirical advances in the field. Multilevel modeling more accurately estimates employee and organizational consequences of HRM practices by taking into account the interactions of organizational contextual factors of higher level variables, as well as including individual factors. This collection includes a number of empirical studies that demonstrate how multilevel modeling is employed in testing hierarchal HRM models. We hope this special issue heralds a new era of multilevel HRM research

Disclosure statement

No potential conflict of interest was reported by the authors.

Acknowledgement

This study was partially supported by Chinese National Science Foundation Grant (71572157).