رهبری و سطوح تجزیه و تحلیل: مرور کیفیت علوم

The purpose of this article is to present a comprehensive, qualitative, narrative review of the leadership literature with an explicit focus on levels-of-analysis issues. Focusing on conceptual and empirical publications (books, book chapters, and journal articles) over the last 10 years in 17 areas of leadership research, we reviewed and coded 348 journal articles and book chapters for the degree of appropriate inclusion and use of levels of analysis in theory formulation, construct/variable measurement, data analytic techniques, and inference drawing. In general, while the literature on leadership is vast and growing, relatively few studies in any of the areas of leadership research have addressed levels-of-analysis issues appropriately in theory, measurement, data analysis, and inference drawing. Nevertheless, the findings reported are encouraging, as levels issues are still relatively new to the leadership field and some progress clearly has been made in the last decade. The implications of the current state-of-the-science for future research and the advancement of study in leadership are discussed.

Levels-of-analysis issues and multiple-level approaches are becoming increasingly important in many areas of organizational research and, in particular, in the literature of leadership. Various scholars (Dansereau et al., 1984*, Dansereau et al., 1999*, House et al., 1995*, Klein et al., 1994* and Rousseau, 1985*) have noted the importance of clearly specifying the level(s) of analysis at which phenomena are expected to exist theoretically, and have stated that it is critical to ensure the measurement of constructs and data analytic techniques correspond to the asserted level(s) of analysis, so that inference drawing is not misleading or artifactual. In the leadership literature, these notions were brought to the forefront of research about two decades ago in the work of Dansereau et al. (1984). Subsequently, when F. Yammarino was Senior Editor of The Leadership Quarterly, he commissioned a two-part special issue (published about one decade ago in 1995), guest edited by F. Dansereau, on 13 multiple-level approaches to leadership. Ultimately, this work culminated in a two-volume research monograph edited by Dansereau & Yammarino, 1998a* and Dansereau & Yammarino, 1998b* on the multiple-level approaches to leadership. In this two-volume set, the editors reprinted the 13 original special issue articles, added the key measurement instruments for each leadership approach, commissioned and published two commentaries on each original article, and published a reply by the original article authors to the commentaries on their work. As this two-volume anthology was, at that time, the state-of-the-science on multiple levels-of-analysis issues and leadership, we use this past work as the starting point for our current effort. In particular, as 10 years have passed since the publication in The Leadership Quarterly of this original multi-level leadership work, we thought it appropriate to consider the current state-of-the-science on leadership and levels-of-analysis issues. Specifically, our purpose here is to provide a comprehensive, qualitative, narrative review of the literature on leadership with an explicit focus on levels-of-analysis issues. Levels of analysis are the entities or objects of study about which we theorize, and are integral parts of the definitions of constructs, operationalizations of measures, and empirical tests of theoretical associations (Dansereau & Yammarino, 2000*, Dansereau & Yammarino, 2003*, Dansereau & Yammarino, 2005*, Yammarino & Dansereau, 2002* and Yammarino & Dansereau, 2004*). In the various areas of leadership research, key levels of analysis are individuals or persons (independent human beings), dyads (two-person groups and interpersonal relationships), groups (work groups and teams), and organizations (collectives larger than groups and groups of groups) (see Dansereau et al., 1984*, Yammarino, 1996* and Yammarino & Bass, 1991*). In the current state-of-the-science review, 348 conceptual and empirical publications (i.e., book chapters and journal articles) in 17 primary areas of leadership research were reviewed and coded in terms of (1) the degree of appropriate inclusion of levels of analysis in theory and hypothesis formulation; (2) the extent to which levels of analysis are represented appropriately in the measurement of constructs and variables; (3) the degree to which levels of analysis are addressed in data analytic techniques; and (4) the extent to which theory and data are aligned from a levels-of-analysis perspective in drawing inferences. The 17 primary approaches to leadership here include the 13 approaches presented in detail in the 1995 two-part special issue in The Leadership Quarterly and in the Dansereau & Yammarino, 1998a* and Dansereau & Yammarino, 1998b* volumes — Ohio State, contingency, participative, charismatic, transformational, leader–member exchange, information processing/implicit, substitutes, romance, self-leadership, multiple linkage, multilevel/leaderplex, and individualized; two additional classical approaches discussed briefly by Dansereau & Yammarino, 1998c* and Dansereau & Yammarino, 1998d* – path-goal and vertical dyad linkage; and two other established approaches – situational and influence tactics. Clearly, several models of leadership worthy of consideration were not involved in the current review. For example, the pragmatic or functional leadership approach (e.g., Mumford, Zaccaro, Harding, Jacobs, & Fleishman, 2000), strategic leadership/upper echelon theory (e.g., Finkelstein & Hambrick, 1996), and shared leadership (e.g., Pearce & Conger, 2003) among others were not included. We chose not to include certain models because of space and time considerations (i.e., the length of manuscript and additional coding time required), but more importantly we tried to align our review with the 1995 special issue in The Leadership Quarterly, the Dansereau & Yammarino, 1998a* and Dansereau & Yammarino, 1998b* volumes, and other “classical” approaches and/or work associated with the authors in the 1995 and 1998 efforts. Ultimately, we have made some choices with which some readers may disagree, but with which we believe most researchers would be comfortable. Conducting a state-of-the-science review and analysis of the 17 selected approaches in the leadership literature seemed important for at least three reasons (beyond merely marking a decade of time since many of the approaches were critically examined in detail). First, given the vast and growing literature on theories and models of leadership, it appears to be an appropriate time to “take stock” of this work. This is especially critical since relatively little of the research to date in these areas of research, as noted below, explicitly focuses on multiple levels-of-analysis issues. Understanding how and if levels are specified permits an examination of the potential or degree of prevalence of theoretical misspecification. Moreover, identification of relevant levels-of-analysis issues may help account for mixed, inconsistent, and contradictory findings in prior research. Second, such a levels-of-analysis examination is critical prior to conducting any comprehensive meta-analysis of theories and models of leadership, which must, at a minimum, account for specific individual-level, within-organization, and organizational-level population parameter estimates (see Ostroff & Harrison, 1999). Without such levels-based efforts, comprehensive meta-analyses cannot be conducted accurately and theoretical advancement is inhibited. Third, only by fully incorporating levels of analysis in theory, measurement, data analysis, and inference drawing can a more comprehensive, integrative, and testable theory of leadership, regardless of approach or realm, result (see Dansereau et al., 1984*, Dansereau & Yammarino, 1998a*, Dansereau & Yammarino, 1998b*, Dansereau & Yammarino, 1998c*, Dansereau & Yammarino, 1998d* and Yammarino, 1996*). Without explicit incorporation of levels-of-analysis issues, incomplete understanding of a construct or phenomena may lead to faulty measures, inappropriate data analytic techniques, and the drawing of erroneous conclusions. As pointed out in the literature on multiple levels of analysis, it is absolutely critical to explicitly specify levels in theory formulation, measurement, data analysis, and inference drawing. Without such a complete and explicit specification of and test for levels of analysis, theory building and theory testing are incomplete and faulty and can lead to erroneous conclusions. For example, Yammarino (1996) demonstrated how theory formulation can be enhanced by incorporating levels of analysis. Moreover, Schriesheim, Castro, & Yammarino (2000) have shown that incorrect conclusions are drawn when researchers rely solely on raw-score-based analyses that ignore levels-of-analysis tests. However, when traditional and multivariate multi-level techniques are employed, inferences become clarified and explicated. Theoretical revolutions in science often occur when other levels of analysis are considered. For example, a revolution in biology occurred when some theorists suggested, and subsequently demonstrated, that evolution can occur at a level of analysis higher than the organism level. Likewise, a well-known revolution in physics occurred when some theorists asserted, and subsequently demonstrated, that quantum mechanics operate at a level of analysis lower than the atomic level. In this same way, leadership theory building can advance when we include lower and higher levels of analysis in theory development and hypothesis generation. In the measurement arena, without explicit consideration of levels of analysis, we do not know to what entities measurements refer. The most reliable measurements we produce are of little or no value if they are not also construct-valid measurements. To be so, such measurements must include the referent (entities or levels of analysis) explicitly. If they do not, a situation develops like some have described for IQ tests: we don't know what they measure, or for whom, but they measure well! In data analysis, lack of the use of appropriate multi-level techniques can lead to statistical artifacts and aberrations. Findings from non-multi-level techniques can indicate “effects” when none actually exist or can “produce effects” at one level when those actually reside at another level of analysis. If theory development, measurement, and/or data analysis fails to address levels-of-analysis issues, there is no way to draw accurate levels-based inferences. The mixing, mismatching, or non-use of levels in any of these three realms limits one's ability to employ a strong inference approach that incorporates multiple levels of analysis. All these issues have been discussed in detail with extensive examples provided in the work of Dansereau et al. (1984), Dansereau & Yammarino (2000), Dansereau et al. (1999), Schriesheim et al. (2000), Schriesheim, Castro, Zhou, & Yammarino (2001), Yammarino (1996), Yammarino & Bass (1991) among others. In fact, Dansereau & Yammarino, 1998c* and Dansereau & Yammarino, 1998d*, the various theorists and researchers in their two-part monograph (see Dansereau & Yammarino, 1998a* and Dansereau & Yammarino, 1998b*), and others (e.g., Avolio & Yammarino, 2002•, Bass, 1990•, Dansereau et al., 1999*, Yammarino, 1996* and Yammarino & Bass, 1991*) have indicated that several alternative levels of analysis are theoretically plausible and empirically viable for numerous approaches to leadership. Consistent with this view, we found (as noted below) multiple levels of analysis were plausible in theory, measurement, data analysis, and inference drawing about the 17 theories and models of leadership investigated here. As such, we sought to complete a comprehensive analysis and coding of these issues for the leadership approaches. In our review, however, we have purposely avoided an assessment of many unpublished papers, unpublished dissertations, and some published articles in what the field might perceive as “hard to locate” or “lesser quality” journals. We have focused our review on major books, significant compendia, and mainstream journal article publications. The search for these publications was facilitated by some major review articles and databases. Moreover, our review begins in the year 1995, the date of the The Leadership Quarterly two-part special issue on multiple-level approaches to leadership. Though selective, we believe our review is quite comprehensive, including 348 publications. In this work, and especially the coding of the articles and book chapters, we were guided by major works on multiple levels-of-analysis issues, following the approach of Dansereau, Yammarino, and colleagues (e.g., Dansereau et al., 1984*, Dansereau & Yammarino, 2000*, Dansereau & Yammarino, 2003*, Dansereau & Yammarino, 2005*, Dansereau et al., 1999*, Yammarino & Dansereau, 2002* and Yammarino & Dansereau, 2004*) and Rousseau (1985). In the remainder of this article, we first highlight levels-of-analysis issues, especially as they apply to the literature on leadership. Then the coding scheme employed, which has been used in at least two prior studies (Dionne et al., 2004* and Yammarino et al., 2002*), is described. Next, for each of the 17 approaches to leadership, we present a brief description of the theory or model involved and then summarize some of the key findings from the conceptual and empirical literature on these leadership approaches from a levels-of-analysis perspective. Finally, we draw out key implications for future theory building and theory testing in the realms of leadership addressed by these 17 approaches.

In terms of the overall results, when combining the assessments and evaluations of levels-of-analysis issues for all criteria, the findings indicate that 19 of 211 empirical publications on the 17 approaches to leadership, or 9%, addressed levels-of-analysis issues appropriately in all four areas of theory and hypothesis formulation, measurement, data analysis, and inference drawing. Thus, in the last decade, the following 19 empirical publications offer encouragement to the field that levels issues are in fact “addressable” and provide a template or “best practices” approach for researchers to follow in their own work: • Berson and Avolio (2004) • Cogliser and Schriesheim (2000) • Dansereau et al., 1995 and Dansereau et al., 1998a • Hofmann, Morgeson, and Gerras (2003) • Howell and Hall-Merenda (1999) • Kark, Shamir, and Chen (2003) • Kim & Yukl, 1995 and Kim & Yukl, 1998 • Schriesheim, Castro, & Yammarino (2000) • Schriesheim, Castro, Zhou, and Yammarino (2001) • Schriesheim et al., 1995 and Schriesheim et al., 1998a • Schriesheim, Neider, and Scandura (1998) • Shamir, Zakay, Breinin, and Popper (1998) • Sivasubramaniam, Murry, Avolio, and Jung (2002) • Sosik, Avolio, and Kahai (1998) • Sosik, Godshalk, and Yammarino (2004) • Sosik, Kahai, and Avolio (1998) • Tosi, Misangyi, Fanelli, Waldman, and Yammarino (2004) • Yammarino, Dubinsky, Comer, and Jolson (1997) • Yammarino, Spangler, and Dubinsky (1998) . In contrast, the remainder of the field of leadership research, as represented by these 17 leadership approaches, seems built upon empirical publications (about 91% of those reviewed here) that do not address adequately levels-of-analysis issues. This assessment is a troubling state-of-the-science, especially given that we have had knowledge of these issues for two decades (e.g., Dansereau et al., 1984* and Rousseau, 1985*) and repeated calls to attend to them over the past decade (e.g., Bass, 1990• and Bass, 2002*). Perhaps, as implied by much of the work reviewed here, leadership approaches are generally an individual leader-style phenomenon and thus this assessment is less bothersome. Nevertheless, even in these cases, alternative levels of analysis need to be “eliminated” theoretically and empirically. Moreover, other constructs (i.e., precursors, consequences, mediators, moderators) in the various theories and models of leadership may operate at other levels of analysis. Again, we believe fuller theoretical specification, measurement, data analysis, and inference drawing, with explicit consideration of levels of analysis for leadership approaches, are required to advance theory building and theory testing. Specifically, to help foster these efforts, we offer the following brief recommendations for conducting multi-level research in all areas of leadership. ▪ In terms of theoretical recommendations: 1. Define the level of analysis of the unit(s) of interest, i.e., the entity (entities) to which theoretical generalizations apply. 2. Define the level of analysis of the associated concepts, constructs, variables, and relationships. 3. Provide a theoretical justification for everything included in #1 and #2 above. 4. Specify the boundary conditions, including and based upon levels of analysis, for everything in #1, #2, and #3 above. ▪ In terms of measurement recommendations: 1. Construct measures at the same level of analysis depicted in the theory, models, and hypotheses. 2. If this is not possible or feasible, employ appropriate aggregation (or disaggregation) techniques and justify the use of these techniques. 3. Further validate a measure, even an established measure, if it has been modified or adapted to account for various or different levels of analysis than originally intended. ▪ In terms of data analytic technique recommendations: 1. Permit theory (variables, relationships, and levels of analysis) to determine the multi-level technique to be used. 2. Employ appropriate multi-level techniques if the “referent(s)” (entities of interest) are at a level of analysis higher than the individual level. ▪ In terms of inference drawing recommendations: 1. Include levels of analysis in both theory (i.e., as the entities) and data (i.e., as the samples and subjects). 2. State which relationships hold across different levels of analysis in terms of multi-level, cross-level, mixed effects, and mixed determinants models. By following these general recommendations and fully and explicitly incorporating multiple levels of analysis in theoretical formulations, measurement, data analysis, and inference drawing, more comprehensive and integrative theories of leadership can be built and tested. We hope this state-of-the-science review of the literature has provided readers with hope. Ignorance of levels-of-analysis issues has hurt prior work on leadership but is providing much needed improvement in current research. Knowledge of levels-of-analysis issues will advance theory building and theory testing in all areas of leadership research. We encourage readers to “lead” the charge in this literature through the use of multiple levels of analysis in theory formulation, measurement, data analysis, and inference drawing.