Training and development as a leverage point

Senge (Reference Senge1990: 114) noted that ‘the bottom line of systems thinking is leverage - seeing where actions and changes in structures can lead to significant, enduring improvements. Often leverage follows the principle of economy of means: where the best results come not from large-scale efforts but from small well-focused actions.’ In systems theory, a leverage point is a place in a system wherein an alteration can lead to a significant change in the functioning of the whole system. A question arises: is there a leverage point in HPWS wherein a percentage change of a constituent part of the system may lead to a significant change in the performance effect of HPWS? To address this question, we need to examine individual constituent parts of HPWS. Thus far, little effort has been made to address this question. Although Bell, Brown, and Weiss (Reference Bell, Brown and Weiss2018) considered team composition to be a leverage point in enhancing human capital, no studies have examined the existence of a leverage point in larger managerial systems like HPWS.

HPWS consists of various constituent parts. HRM researchers usually took two approaches to decompose HPWS. The first approach distinguishes between several subsystems of HPWS. The AMO framework is a typical example of this approach (Appelbaum, Bailey, Berg, & Kallegerg, Reference Appelbaum, Bailey, Berg and Kallegerg2000; Chowhan, Pries, & Mann, Reference Chowhan, Pries and Mann2017). The second approach, often called the common practices model, focuses on individual HRM practices commonly adopted in real-world businesses. Rabl et al.'s (Reference Rabl, Jayasinghe, Gerhart and Kühlmann2014) framework of six HPWS components or practices is a typical example of this approach. We contend that the second approach is more appropriate for leverage point analysis since it allows us to anchor on a specific practice (rather than a large subcluster of multiple practices) that is central to amplifying the synergy of HPWS.

The six components of HPWS in Rabl et al. (Reference Rabl, Jayasinghe, Gerhart and Kühlmann2014)'s framework include compensation, employee relations, performance management, T&D, promotion, and recruitment and selection that are widely discussed in HRM research. We contend that OP is determined by employees' ability, motivation, and opportunities (Appelbaum et al., Reference Appelbaum, Bailey, Berg and Kallegerg2000). All the six components of HPWS interact with one another to enhance AMO and, through them, OP. However, the degree to which these practices take effect on performance outcomes varies substantially.

Among the six components, we believe, T&D is likely to be a leverage point of HPWS. That is, a small percentage change of T&D may lead to a significant amount of change in the overall functionality of HPWS in terms of its effect on performance improvement in an organization. Relatively speaking, T&D is less dependent on the rest five components than they are dependent on T&D. This is because T&D simultaneously influences all three subsystems (AMO) of HPWS.

T&D has long been recognized as an ability-enhancing activity (Hong, Jiang, Liao, & Sturman, Reference Hong, Jiang, Liao and Sturman2016; Jiang, Lepak, Hu, & Baer, Reference Jiang, Lepak, Hu and Baer2012). It enhances employee knowledge, skills, and abilities to fulfil immediate and long-term job requirements (Sum, Reference Sum2009). With firm-specific T&D programs, an organization can cultivate employee capabilities in line with its strategic directions and thus enhance its competitive advantages (Vazquez-Bustelo & Avella, Reference Vazquez-Bustelo and Avella2019). It also helps employees convert new knowledge and skills into competencies to aid organizational ambidexterity by which the organization exploits current business strengths while exploring future capability potential for sustainable gains (Mom, Chang, Cholakova, & Jansen, Reference Mom, Chang, Cholakova and Jansen2019).

Scholars suggested that T&D also enhances employees' motivation. It delivers a signal that the organization values its employees and is willing to help them develop personal abilities and qualities (Tai, Reference Tai2006). It also shows an organization's commitment to human capital development, enhancing employees' self-efficacy and self-actualization (Demirbag, Collings, Tatoglu, Mellahi, & Wood, Reference Demirbag, Collings, Tatoglu, Mellahi and Wood2014).

In the prior literature, T&D has been perceived as an instrument increasing opportunities for employees to contribute to the organization. With enhanced abilities and motivation, employees can play a more influential role in an organization. They can have more chances to participate in, for instance, decision-making, teamwork, information sharing, network building, and job design (Chowhan, Reference Chowhan2016; Sum, Reference Sum2009).

Our arguments for the vital importance of T&D are consistent with empirical findings. Jennings, Cyr, and Moore (Reference Jennings, Cyr, Moore, Moore and Jennings1995) found that managers considered T&D as the most important strategic HRM practice. Darwish, Singh, and Wood (Reference Darwish, Singh and Wood2016) also found evidence that among all HRM practices examined, T&D was the only one that delivered significantly positive impact on the performance of companies. Likewise, in a meta-analysis, Tzabbar, Tzafrir, and Baruch (Reference Tzabbar, Tzafrir and Baruch2017) ascertained that T&D is a universally best HRM practice for performance improvement across industry sectors, organizational sizes, and societal settings. Based on these prior arguments and evidence, we propose the following hypothesis.

The role of country of origin

A leverage point takes shape in a system under certain conditions (Senge, Reference Senge1990). T&D may serve as a leverage point in HPWS contingent on several factors. Country of origin is likely to be such a contingency. The concepts of HRM and HPWS originated in the US and then spread to other advanced countries in Western Europe, Canada, Australia, and New Zealand which share institutions, level of economic development, ideology, and management philosophy similar to those in the US (Drucker, Reference Drucker1954; Singh, Darwish, & Potočnik, Reference Singh, Darwish and Potočnik2016). It was not until recently that these concepts were introduced to developing countries. For example, modern philosophy and practices of HRM were applied in Chinese organizations only after the middle and late 1990s, before which people management function mainly focused on the routine personnel administration (Zhao & Du, Reference Zhao and Du2012). Development of HPWS is still at an early stage in many developing countries. As compared to firms in developing countries, firms in advanced countries have accumulated rich experience in implementing HPWS (Zhai & Tian, Reference Zhai and Tian2020).

The variation in experience in implementing HPWS bears enormous significance to the role of T&D as a leverage point in the system. To begin with, it affects the generation, acquisition, and utilization of new information and knowledge through T&D that helps enhance employees' capacity for implementing other HPWS practices (Lepak & Shaw, Reference Lepak and Shaw2008). In organizations with a high level of such experience, employees, who have been involved in HPWS routines for long, are more capable of effectively participating in T&D to maximize the functionality of HPWS (Ramsay, Scholarios, & Harley, Reference Ramsay, Scholarios and Harley2000). In organizations with a low level of such experience, by contrast, employees lack exposure to HPWS routines, including T&D. They usually have limited opportunities and resources to practice their skills in utilizing T&D instruments, and thus are relatively incompetent to perform HPWS and other job tasks effectively (Bos-Nehles, Van Riemsdijk, & Looise, Reference Bos-Nehles, Van Riemsdijk and Looise2013). In addition, these organizations usually do not have adequate capacity to ensure expected T&D quality and quantity (Garavan, McCarthy, Lai, Murphy, Sheehan, & Carbery, Reference Garavan, McCarthy, Lai, Murphy, Sheehan and Carbery2021). It is difficult for employees in these organizations to leverage T&D investment to achieve the same performance outcomes as their counterparts in organizations with a high level of such experience (Wang & Chen, Reference Wang and Chen2013).

Accumulated experience in implementing HPWS not only helps an organization generate new information, but also enhances its ability to absorb available knowledge and information from the external environment to strengthen the performance impact of HPWS (Han et al., Reference Han, Kang, Oh, Kehoe and Lepak2019). Absorptive capacity refers to ‘the ability of a firm to recognize the value of new, external information, assimilate it, and apply it to commercial ends’ (Cohen & Levinthal, Reference Cohen and Levinthal1990: 128). A high level of experience in implementing HPWS indicates a high level of capacity to absorb useful information from the external environment. HPWS and OP are connected by an open-loop mechanism (Shin & Konrad, Reference Shin and Konrad2017). It means that in developing and implementing HPWS, an organization has to respond to feedback from the external environment on the performance effect of HPWS. The backward-going feedback from the external environment need to be correctly understood, adequately assimilated, and properly applied in business operations to achieve commercial ends. A high level of absorptive capability strengthens the critical role of T&D in HPWS, resulting in an upward spiral of improvement in OP. Based on these prior arguments, we propose the following hypothesis.

The role of industrial affiliation

Another contingency factor is industrial affiliation. Scholars have discussed the difference in the effect of HPWS on OP between manufacturing and service industries (Combs et al., Reference Combs, Liu, Hall and Ketchen2006). To date, however, we know little about the difference in the degree to which T&D can serve as a leverage point to influence the functioning of HPWS between manufacturing and service industries. Service industries differ significantly from manufacturing industries. The differences between the two industrial sectors may affect the role of T&D as a leverage point in HPWS.

Some prior studies suggested that HPWS is better aligned with manufacturing industries than service industries, and that T&D is an effective approach to enhancing employee ability to work in manufacturing firms (Bowen, Reference Bowen1986). This is because employees in manufacturing firms need more practical skills in operating machinery and adapting to changes in physical infrastructure upgrading, which can be achieved through T&D relatively easily (Evans & Davis, Reference Evans and Davis2005). Also, some elements of HPWS, such as teamwork, are better aligned with manufacturing industries than service industries, and T&D is an effective instrument to enhance employees' ability to work in a team (Jennings, Cyr, & Moore, Reference Jennings, Cyr, Moore, Moore and Jennings1995).

We argue, however, that the opposite might be true for two reasons. First, a key function of HPWS is to enhance employees' discretion over what and how they perform their roles (Liao, Toya, Lepak, & Hong, Reference Liao, Toya, Lepak and Hong2009). Employees in service industries need greater discretion than their counterparts in manufacturing industries because they often work individually on their posts (Rosenthal, Hill, & Peccei, Reference Rosenthal, Hill and Peccei1997). T&D plays an important role in enhancing employees' ability to make correct and prompt decisions at their discretion (Darwish, Singh, & Wood, Reference Darwish, Singh and Wood2016). Second, HPWS influences not only employees, but also the interaction between employees with customers, and employees in service industries are closer to customers than their counterparts in manufacturing industries (Datta, Guthrie, & Wright, Reference Datta, Guthrie and Wright2005). Appropriate T&D program provides employees with the abilities and skills to interact with customers, to understand feedback from customers, and to absorb the new information generated from customer feedback (Darwish, Singh, & Wood, Reference Darwish, Singh and Wood2016). Accordingly, T&D should be more likely to serve as a leverage point in service than manufacturing industries. We propose the following hypothesis.

Sample

We aimed to include all published and unpublished studies on the HPWS–OP relationship available by December 2021 to establish an unbiased and exhaustive sample pool for our meta-analysis (Card, Reference Card2012). Keywords used for sample search included ‘high-performance work system,’ ‘organizational performance,’ and their variations and combinations. We used multiple databases and search engines, screened major journals and leading conference programs in the HRM discipline for relevant articles. We further scanned the reference lists of previous meta-analyses (e.g., Rabl et al., Reference Rabl, Jayasinghe, Gerhart and Kühlmann2014; Rauch & Hatak, Reference Rauch and Hatak2016) for possible missing articles. We also contacted the corresponding authors for articles that are irretrievable or not downloadable online. Details of the sample search procedure are presented in Appendix 1.

To be included in our meta-analysis, a study must report (1) HPWS–OP correlations or other statistics that can be converted to an effect size; (2) measures of either financial or operational OP at the firm or establishment level, regardless of performance measured by objective (e.g., accounting figures from corporate annual reports or financial statements) or subjective indices (e.g., perceived OP or perceived market positioning over competitors); (3) the aggregate effect of HPWS or HR systems instead of individual HR practices; (4) composition of the practices included in HPWS; (5) at least one of the six practices of HPWS (i.e., the study was excluded if it did not include any of the six practices of HPWS which are of the interest of this meta-analysis); and (6) the country and industry from which data were collected (Combs et al., Reference Combs, Liu, Hall and Ketchen2006; Rabl et al., Reference Rabl, Jayasinghe, Gerhart and Kühlmann2014).

Following these criteria, we retrieved 236 studies, representing 240 independent samples of 59,207 firms and establishments from 35 countries and regions. A list of these studies is available in the Supplementary files of this study.

Effect size

We coded the effect size in each study by Pearson's correlation r. For studies that reported multiple dimensional correlations of HPWS and OP, or a particular HPWS and OP construct over several time slots (e.g., Chadwick, Super, & Kwon, Reference Chadwick, Super and Kwon2015), a linear composite correlation was computed to estimate an aggregated effect size.

We then used the sample size of each study to correct for sampling errors and inter-item reliability (Cronbach's α) to correct for measurement errors. As for inter-rater reliability of the coding task, one of the authors performed data coding tasks independently, and the other author crosschecked 33% of the coding outcomes. Since both authors have reached agreements in 95% of the coding outcomes and resolved all disagreements through subsequent discussions, the accuracy and internal consistency of the coding work were satisfactory (McHugh, Reference McHugh2012).

HPWS composition

As discussed earlier, we focus on the impact of a percentage change of each HPWS component in the common practice model on the HPWS–OP relationship. We thus need to calculate the percentage composition of the six components in HPWS. We followed Rabl et al. (Reference Rabl, Jayasinghe, Gerhart and Kühlmann2014) to decompose the HPWS construct. Specifically, we calculated the percentage of each HPWS component that fell into the six practices of interest according to the construct measurement reported in sample studies. The sum of these percentages does not necessarily equal to 100% since HPWS components were highly divergent and overlapping across sample studies, indicating a lack of consensus in measuring benchmark HPWS components among HRM scholars (Vazquez-Bustelo & Avella, Reference Vazquez-Bustelo and Avella2019). Percentages of the six HPWS components in the sample studies are presented in Appendix 2 of this study. In the whole sample, as shown in Figure 1, the percentage was on average 16% for compensation, 24% for employee relations, 12% for performance management, 18% for T&D, 6% for promotion, and 11% for recruitment and selection.

Figure 1.

Average percentage of the HPWS components (whole sample).

Control variables

We included four control variables which were considered to influence the HPWS–OP relationship in HRM literature. The first was country of origin. We followed the World Economic Outlook's (International Monetary Fund, 2021) classification to construct a categorical variable with 1 denoting studies using samples from advanced countries and zero denoting studies using samples from developing countries. In the whole sample set, 179 studies were conducted in advanced countries and 61 in developing countries.

Another control variable is performance measures, which was categorical. This variable was coded as 1 when OP was measured in financial terms such as returns on asset ratio, Tobin's Q, and market shares, and zero when OP was measured in operational terms such as innovation capability, operational efficiency, and productivity. Among the 240 samples, 66 used financial performance measures whereas 174 used operational performance measures.

Firms and establishments included in sample studies fell into three industry categories: manufacturing, service, and mixed industries. We constructed two categorical variables to control for the industrial affiliation for sample studies. Of the 240 samples, 51 were in manufacturing industries, 56 in service industries, and the remaining133 were in mixed industries.

We also followed Combs et al.'s (Reference Combs, Liu, Hall and Ketchen2006) suggestion to control for the potential influence of level of analysis. The variable was coded as 1 if the analysis in a sample study was conducted at the entire firm or organizational level covering multiple establishments, and zero if the sample in a study was collected at the establishment level, such as a department, team, or branch in a single location within a large enterprise. Among the 240 samples, 185 were in the firm-level category and 55 were in the establishment-level category.

Estimation strategy

We performed empirical assessments mainly using the ‘metafor’ package of R program 3.3.3 (Viechtbauer, Reference Viechtbauer2010). We normalized HPWS–OP effect sizes by Fisher's z (Lipsey & Wilson, Reference Lipsey and Wilson2001), and used the sample size weighting method to estimate the average HPWS-OP effect size. We used the average reliabilities of HPWS constructs r_xx = .819 and OP constructs r_yy = .833 for correcting measurement errors, and computed a weighted mean true score correlation ρ (Hunter & Schmidt, Reference Hunter and Schmidt2004).

As the six HPWS components, the core focus of our study, are calculated as percentages rather than dummies, we could not conduct heterogeneity test of subgroup analysis as was often undertaken in ordinary meta-analysis. Instead, we carried out meta-regressions.

Whole sample meta-analysis

Table 1 reports the descriptive statistics and correlation coefficients of the variables in the whole sample. The unweighted mean of observed HPWS–OP effect sizes, r, was .297. The coefficient of the correlation between the effect size r and T&D was positive and statistically significant (R = .130, p < .05) whereas the coefficients of the correlation between the effect size r and the rest five HPWS components were all statistically indifferent from zero. The results seemed to indicate that the HPWS–OP relationship is related to the percentage change of T&D, but not to that of other HPWS components.

Table 1.

Descriptive statistics and correlations of variables of the whole sample^a

^a Missing observations removed by pairwise deletion in correlation computation.

^b k = number of studies.

^c N = total sample size of the k studies.

^d Unweighted mean (arithmetic average) of observed effect sizes.

*p < .05; **p < .01; ***p < .001.

As for control variables, the coefficient of the correlation between the effect size r and advanced country was negative and statistically significant (R = −.244, p < .001), as was the coefficient of the correlation between the effect size r and the financial performance (R = −.236, p < .001). However, the coefficients of the correlation between the effect size r and manufacturing industry, service industry, and firm-level analysis were all statistically indifferent from zero. The results seemed to indicate that the HPWS–OP relationship was likely influenced by country of origin and performance measures.

Table 2 reports the meta-regression analysis output. We first ran model 1 in which only control variables were included. The coefficient of advanced country was negative and statistically significant (β = −.113; p < .001), as was the coefficient of financial performance (β = −.115; p < .001). However, the coefficients of manufacturing industry, service industry, and firm-level analysis were all statistically indifferent from zero. The results were consistent with correlation analysis, suggesting that country of origin and performance measures influenced the HPWS–OP relationship, but industrial affiliation and level of analysis did not have any significant impact on the relationship in the whole sample.

Table 2.

Results of meta-regression analyses of the whole sample^a

^aWeighted least square regression by inverse variance weighted variables in a random-effect model was applied following the meta-regression approach suggested by Lipsey and Wilson (Reference Lipsey and Wilson2001).

^bStandard errors are in parentheses.

^cT ²: restricted maximum likelihood (REML) estimate of between-study variance.

^dI ²: proportion of variation due to heterogeneity in residual.

^eR ²: proportion of between-study variance explained.

*p < .05; **p < .01; ***p < .001.

We then ran model 2 in which the percentages of the six HPWS components were introduced. The signs of all the control variables remained unchanged. The coefficient of T&D was positive and statistically significant (β = .302; p < .05) whereas the coefficients of all the remaining five HPWS components were statistically indifferent from zero. The results were consistent with correlation analysis, suggesting that among all the six HPWS components, only T&D moderated the HPWS relationship. The results fully supported hypothesis 1, indicating that the HPWS–OP relationship was highly sensitive to the percentage change of the T&D component in HPWS, but not on the percentage change of any of the other five HPWS components. T&D served as a leverage point in HPWS.

Subsample meta-analysis

Advanced versus developing countries

To test hypothesis 2, we divided the whole sample into two subsamples, the developing country subsample and the advanced country subsample, to examine how country of origin influences the functioning of T&D as a leverage point in HPWS. We ran meta-regressions for the two subsamples, respectively, and reported the output in Table 3.

Table 3.

Results of meta-regression analyses in subgroup samples: country of origin^a

^aWeighted least square regression by inverse variance weighted variables in a random-effect model was applied following the meta-regression approach suggested by Lipsey and Wilson (Reference Lipsey and Wilson2001).

^bStandard errors are in parentheses.

^cT ²: restricted maximum likelihood (REML) estimate of between-study variance.

^dI ²: proportion of variation due to heterogeneity in residual.

^eR ²: proportion of between-study variance explained.

*p < .05; **p < .01; ***p < .001.

The coefficient of T&D was positive and statistically significant (β = .374; p < .05) in model 4 in the advanced country subsample, but statistically indifferent from zero in model 6 in the developing country subsample. Moreover, the coefficients of all the remaining five HPWS components were statistically indifferent from zero in both subsamples. The results supported hypothesis 2, indicating that the positive HPWS–OP relationship is highly sensitive to the percentage change of the T&D component in HPWS in advanced countries rather than developing countries. T&D served as a leverage point in advanced countries, but not in developing countries.

As for control variables, in both subsamples, the coefficient of financial performance was negative and statistically significant (β = −.121; p < .001 in model 4, and β = −.178; p < .05 in model 6) whereas the coefficients of all the remaining control variables were statistically indifferent from zero after the six HPWS components were introduced in models 4 and 6. The results suggested that performance measures influenced the HPWS–OP relationship in both developing and advanced countries.

Manufacturing versus service industry

To test hypothesis 3, we divided the whole sample into two subsamples, the manufacturing industry subsample and the service industry subsample, to examine how industrial affiliation may influence the functioning of T&D as a leverage point in HPWS. We ran meta-regressions for the two subsamples, respectively, and reported the output in Table 4.

Table 4.

Results of meta-regression analyses in subgroup samples: industrial affiliation^a

^aWeighted least square regression by inverse variance weighted variables in a random-effect model was applied following the meta-regression approach suggested by Lipsey and Wilson (Reference Lipsey and Wilson2001).

^bStandard errors are in parentheses.

^cT ²: restricted maximum likelihood (REML) estimate of between-study variance.

^dI ²: proportion of variation due to heterogeneity in residual.

^eR ²: proportion of between-study variance explained.

*p < .05; **p < .01; ***p < .001.

The coefficient of T&D was positive and statistically significant (β = .971; p < .05) in model 10 in the service industry subsample, but statistically indifferent from zero in model 8 in the manufacturing industry subsample. Moreover, the coefficients of all the remaining five HPWS components were statistically indifferent from zero in both subsamples. The results supported hypothesis 3, indicating that the positive HPWS–OP relationship is highly sensitive to the percentage change of the T&D component in HPWS in service industries rather than manufacturing industries. T&D served as a leverage point in HPWS in service industries, but not in manufacturing industries.

Regarding control variables, in the manufacturing industry subsample, the coefficient of financial performance was negative and statistically significant (β = −.155; p < .01) whereas the coefficients of all the remaining control variables were statistically indifferent from zero after the six HPWS components were introduced in model 8. The results suggested that performance measures influenced the HPWS–OP relationship in manufacturing industries. In the service industry subsample, by contrast, the coefficients of all the control variables were statistically indifferent from zero after the six HPWS components were introduced in model 10, indicating that none of the control variables influenced the HPWS–OP relationship in service industries.