多層次管理研究：分析層次的概念、理論和方法

Multilevel Research in Management: Conceptual, Theoretical, and Methodological Issues in Level of Analysis

在組織的場域中，個人的行為與態度可能會受到組織環境的影響；同樣的，組織行動也有可能會受到個人因素的作用，因此，研究者在建構與探討組織現象的時候，不能不注意到這個基本的前提。雖然早在1930年代前後，學者在理論建構時便已經注意這一個多層次議題，然而，在實徵研究上，卻常常忽略了組織多層次鑲嵌的本質，仍採取單一分析層次的作法。如此一來，理論思維與實徵策略的不一致，使得我們在組織知識的累積上受到層次謬誤的干擾。大約近十年，管理領域的學者多已接受了組織現象是宏觀因素與微觀因素相互影響的多層次現象，並將此觀念應用於實徵研究。但整體來看，多層次分析仍屬相對少數，國外學界如此，國內學界尤然。如果多層次的現象從單一層次角度切入，最明顯的缺點是可能遺漏了重要的解釋變項，導致解讀偏誤，最嚴重的後果則是知識錯誤的累積。雖然目前國內已經出現以多層次方法處理組織現象的實徵研究，顯示國內管理學者開始對多層次研究產生興趣，但目前卻沒有對於多層次研究在概念、理論、與方法等議題方面深入討論的文章。本文的目的在於整理與評述過去數十年來西方學者在有關多層次研究的概念、理論和方法的發展，並提出我們的補充。除了提出我們的觀察與見解外，文內並舉出一個具體的應用實例，說明不同的分析模式將會產生不同的結果，以此做為學者的參考，俾利在培養正確的概念之後，可以有效地進行多層次研究。

Since around the 1930s, prominent scholars in social psychology and sociology time and again appealed for understanding human behaviors from multiple angles to consider structural factors in the environment besides human characteristics. Taking a similar perspective, management scholars gradually recognized that organizational phenomena are inherently multilevel in nature such that behaviors at any level are very likely a result of elements at various levels interacting in various forms. For example, it is now commonly accepted that individual cognitions are influenced at least in part by the context in which they work and departmental operations are affected by organizational strategy. Similarly, actions at the organizational level such as alliance and merger will not take place unless key persons in the organization (e.g., the CEO or the top management team) perceive such a need and determine to put the need into action. Nonetheless, the empirical studies in organizational science today are mostly single-level analyses, reflecting the fact that research practice lags behind concepts and theories in the management community. Understanding multilevel phenomenon from a single level will result in biased interpretation, and worse incorrect knowledge accumulation. The interest in multilevel research has just started in Taiwan. The objectives of this paper are twofold. First, we want to familiarize readers interested in multilevel research and/or in level of analysis issues with concepts, theories, and methodologies pertaining to multilevel research perspective. It is our hope that this paper would provide readers with preliminary guidance when they feel the need to quickly get hold of those issues. Second, we want to make this study resourceful to our readers in such a way that, from this paper, they find it relatively convenient to locate pioneering and/or influential works as well as statistical packages in multilevel enquiry. We began by elaborating issues dating back in the late 1930s with regard to the fallacies resulted from level misspecifications. While ecological fallacy is committed when researchers interpret macro level correlations as micro level relationships, atomistic fallacy occurs when researchers use micro level correlations to explain macro level phenomena. Organizational scientists have most commonly committed this later fallacy. This section includes our opinions on these wrong level fallacies and their implications. Next, we summarized and commented on the development of multilevel concepts, theories, and analytic techniques over the last several decades in the West. This study identified two lines of clues: conceptual and analytical. Conceptual development involves the process with which researchers with a single level mindset gradually turn to accepting multilevel thinking. It also sketches how researchers believe multilevel model should be logically constructed and how lower level data can be aggregated to upper level. Pioneers in this stream include Denise Rousseau and Williams Glick. The other clue involves the development of analytical criteria of the conditions under which we determine data aggregation is theoretically sound and statistically robust. Lawrence James and Fred Dansereau, among others, are major contributors in this regard. We argued that it takes both types of development to drive the emerging paradigm shift moving from single to multilevel thinking. Citing some important works in the literature, we then address issues with regard to multilevel model construction by elaborating three types of models (single-level, cross-level, homogeneous multilevel) as well as four types of constructs (global, configural, shared, and formative). As we addressed the distinctive characteristics of these model and constructs, we provide guidelines for data collection, data conversions and validation. It is stressed that a good understanding of these issues is essential in making appropriate model specification, which in turn is a building block for sound research design and logical data interpretation to avoid potential fallacies. In the section that follows, we presented an empirical study employing three approaches to examine the relationship between psychological contract breach and objective performance with group cohesiveness as a moderator: individual, group level (both of them single level analyses) and cross level. The results not only demonstrate a case where model misspecification leads to wrong conclusions and exemplify how the organizational phenomenon of interest can be dealt with by research design and data analysis that are explicitly multilevel. In the concluding section, we emphasized again the importance of multilevel thinking in understanding organizational phenomena. Stressing that theoretical basis dictates research design and data analysis, we argued the popular software HLM is not equivalent to MLR. It is our hope that this paper will encourage more serious organizational scholars to devote their future efforts in MLR.