The Application of Latent Class Analysis on the Measurement of HRM Practices

submitted to the DRUID Summer conference 2003 on Creating, Sharing, and Transferring Knowledge: The role of Geography, Institutions, Organisations. The proposed paper in related to conference themes B and E. Exploiting old and new knowledge in combination in order to generate new knowledge is a key feature in achieving competitive advantage of the firm. The emergence of the knowledgebased economy with globalisation, deregulation of markets, changing demands and shorter product lifecycles, has become the norm for most organisations. In order to be competitive, organisations must be able to continually improve their performance by reducing costs, innovating in terms of new products and processes, improving quality, productivity, and speed to the market. According to the resource-based view of the firm (Penrose, 1959; Barney & Link, 1991), competitive advantage can be developed and sustained by creating value in a way that is rare and difficult for competitors to imitate. The resource-based view of the firm argues that traditional sources of competitive advantage like natural resources, technology and economies of scale, are sources that are increasingly easy to imitate, especially in comparison to complex social structures like the employment system within the organisation. In this case human resource strategies in terms of human resource management practices (HRM) may be an important source of sustained competitive advantage (Lado & Wilson, 1994; Pfeffer, 1994). According to Barney (1991) causal ambiguity and path dependency are two of the key factors for making HRM practices difficult to imitate. First, it may be difficult to understand the precise mechanisms in HRM practices. Second, HRM practices are often a complex system of which the value is determined by the interplay of HRM practices and firm policies. Understanding these systems is an organisational capability that is spread across department and people in the firm. HRM systems are moreover path dependent since they often have been developed over time and cannot be purchased in the market by competitors. Even if they could, the management of competing firms may have difficulties in replicating the socially complex elements like culture and interpersonal relationships. Consequently, several scholars have proposed and found that application of HRM practices contributes to firm performance. Much of the research has focused on a single HRM practice, e.g. Mahoney (1995) and Huselid (1995), and it is only recently the link between systems of HRM practices and different types of firm performance has been examined empirically e.g. (Baron & Kreps, 1999; Ichniowski et al., 1997; Mendelson & Pillai, 1999; Michie & Sheehan, 2003) building on the theoretical concept of “Edgeworth complementarity” introduced by Milgrom and Roberts (1990) and Holmstrom and Milgrom (1994). But empirically it is a complicated task to identify complementarity. This may be illustrated by considering the case where we have dichotomous observations on two HRM related variables. In a regression framework complementarity would correspond to a strictly positive coefficient on the cross term of the two variables. However, in the case of e.g. seven dichotomous variables, we would end up estimating a regression with 128 parameters. This is clearly not a feasible solution. Notice that no direct measure of HRM exists and therefore researchers typically use several indicators of the concept. According to Athey and Stern (1998), two types of approaches have been used in the literature. The first relies on testing whether the practices are positively correlated, conditional on observables, as for instance in Laursen and Mahnke (2001). However, no measure of performance is included in their article, and hence complementarity in the usual sense is not tested. The second approach uses various regression techniques in order to measure the productivity effect on a set of HRM variables and their interaction effects. For instance, Ichniowski et al. (1997) finds by applying the second method, a number of HRM systems by using recent literature and bivariate correlations and argues by OLS estimation, that systems of HRM practices determine productivity and quality, while marginal changes in individual work practices have little effect. This approach suffers from the obvious limitation, that the researcher has to specify in advance the exact combinations of HRM practices to be included in the analysis. I.e. there is a very large extent of subjectivity involved both with respect to the number of HRM practices considered as well as the particular combinations used. In the present paper we introduce the latent class analysis as an alternative to the above mentioned methods. Using latent class analysis we are able to generate one measure of the amount of HRM practices employed in each firm. This approach has a number of advantages: First, it takes a systematic approach with respect to the generation of the particular combination of HRM practices. Second, it is a method that takes as a point of departure the often prevailing categorical nature of the measurements of HRM practices into account. Finally, the method offers a statistical test of the validity of the model. The purpose of this paper is first, to argue the advantages of using latent class models on the use and finding of systems of HRM practices and secondly on a empirical sound basis, to document the presence of complementarities of HRM practices. This is implemented by applying a Danish survey on organisation, employee skills and development of new products conducted in 2001 covering 2007 firms in the manufacturing and service sector in the period of 1998-2000.