Limitations and Suggestions for Future Research Like all research, this study has a number of limitations. A limitation of this study is that its sample provides a restricted basis for generaliz- ing our findings. This study focused on firms in two industries in Jordan. The focal setting may limit the generalizability of the findings to or- ganizations within this national context. In ad- dition, the sample size was relatively small and we investigated moderated relationships. Both the small size of our study sample and the focus on moderation increase the risk that significant relationships will not be detected (McClelland & Judd, 1993). However, we received good sup- port for all of our hypotheses and our overall model. Given that the detection of moderation is particularly constrained by small sample size (Cohen, 1988), the results indicate some ro- bustness in our findings. A second limitation of this study, frequently observed in survey-based research, is bias asso- ciated with common method variance (CMV) (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003). We acknowledge that Gerhart, Wright, Mahan, and Snell (2000) highlighted the risks associated with single-source methods of data collection. These authors also state, however, that bias is more likely in large firms in which practices and human resource systems are likely to vary considerably. While their mean organi- zational size was 40,000 employees, the data in the current study was generated from smaller organizations, the majority of which employed between 100 and 300 people. There is support for the argument that smaller firms are unlikely to feature significant within-firm variation in practices and systems, which is likely to improve reliability (Huselid & Becker, 2000). In addition, previous research indicates that CMV is less of a concern for studies with a moderation effect (Schriesheim & DeNisi, 1981). The logic is that common method bias should be present regardless of the level of the moderator. It follows that when the relationship between the predictor, mediator, and dependent construct varies depending on the moderator, as in our study, responses are unlikely to be the result of using the same method—participants have not rated all items similarly as is evident when self-report CMV threatens data validity (Schriesheim & DeNisi, 1981). Moderation models are also less likely to risk bias from participants’ the- orizing about predicted relationships. This risk is more likely in simple linear models (Harrison, McLaughlin, & Coalter, 1996). One approach advocated to minimize com- mon method bias is to specify relationships among the dependent and predictor variables that are complex and therefore not likely to be part of the individual raters’ cognitive maps. Our moderated mediation model is, therefore, likely to reduce the risk of bias asso- ciated with respondents’ theorizing, because such a complex relationship is not likely to be part of the respondents’ theory in use (S. J. Chang, Van Witteloostuijn, & Eden, 2010). Finally, to address the issue of potential common method bias empirically, we used a CMV marker variable (Lindell & Whitney, 2001; Richardson et al., 2009). Analysis using the CMV marker variable indicated that the partial correlation coefficients for all predic- tor and dependent variables remained statis- tically significant ( p = .00 for all correlations) when the marker variable was controlled. In addition, we used a Harman one-factor test (Podsakoff & Organ, 1986). All scale items used to measure the predictor variables were entered into a principal components factor analysis, and the results of the unrotated solution were explored. Four factors were identified with an eigenvalue greater than 1, and no single fac- tor accounted for the majority of variance. This suggests that common method variance is not a significant issue in this study. If com- mon method variance is a significant issue, it is likely that a single factor would emerge or one general factor would account for a major- ity of variance (Posdakoff & Organ, 1986). While the risk of bias is acknowledged and has been addressed by the authors, we also recognize that there is a growing acceptance of research indicating that CMV is unlikely to invalidate study findings, particularly in the investigation of complex models (Crampton & Wagner, 1994; Doty & Glick, 1998; Spector, 2006). We acknowledge, however, that self- report methods may be considered a limita- tion to this study. Our research focused on the frequently utilized outcome measure of organiza- tional financial performance. While this is a robust measure of managerial importance, the relationship between the HR strategic role, HPHRPs, and organizational outcomes beyond financial performance is also war- ranted—for example, investigating the impact of the HR strategic role on employee- level performance such as employee satisfac- tion and voluntary turnover. The potential effect of different HR roles, such as change agent, employee champion, and adminis- trative expert, on organization performance also merits further exploration, as are the potential contributions of each role to orga- nizational performance. Finally, while we investigated the effect of devolvement as an important con- tingency factor in the relationship between the HR strategic role and the adoption of HPHRPs, there are other important contingencies that may influence this relation- ship, such as top management support and IT support. Further research should thus investigate other contingencies affecting this relationship. Despite these limitations, the findings reported here suggest that there is important research potential in investigating the role of coupling of an HR strategic role with the devolvement of respon- sibility for HR execution to line management. This study also indi- cates that our understanding of the impact of the HR strategic role and HPHRPs is significantly enhanced by investigating complex mediat- ing pathways, and also reinforces the merit of investigating relation- ship contingencies. Furthermore, the value of undertaking this research in Jordan is considerable. The findings provided evidence for the applicability of theory and previous empirical work in a non- Western context, particularly the Jordanian context.