IIn the family of management related research, scholars try to understand phenomena in and around organizations, such as employees' perceived justice, patterns of intra-organizational interactions, network relationships of organizations, and consumer responses to brand names. In pursuing these enquiries, they often use self-report instruments to collect data from research subjects. If only one type of survey questionnaire is administered to a single source of respondents and the questionnaire contains both the antecedents and outcome variables, then it is very likely that this research suffers a methodological problem termed common method variance (CMV). CMV will inadequately inflate the relationship between variables, resulting in an increase of statistical significance. Based on such significance, hypothesis is often misjudged as being supported and thus Type I error occurs. Therefore, CMV is regarded as an obvious threat to internal validity. The purpose of this study is to appeal for readers' attention to the CMV problem and to handle it more effectively in future research programs.
To begin with, we articulate what CMV is as well as its causes and impacts. Results from psychometric measurement can be dichotomized as random error variance and systematic variance. This later category in turn consists of two parts: trait variance and method variance (i.e., CMV). Trait variance is the variance reflecting the trait (i.e., construct) measured from a particular sample. Hence, the larger the trait variance is, the higher the construct validity of the particular trait can be. In contrast, both random error variance and method variance are measurement errors. They differ in the fact that method variance, like trait variance, is systematic. Method variance consistently goes along with trait variance and is therefore difficult to detect. Logically, CMV is the part of variance that is totally undesirable and needs to be minimized. Major causes of CMV include the use of same methods (e.g., self-report questionnaire), collecting data from a single source and/or at the same time, respondents' response set, consistency motive, and psychological state (e.g., social desirability, negative affectivity), and other contaminating factors. CMV imposes a negative impact on construct validity, which may lead to misleading statistical significance and eventually inadequate accumulation of management knowledge.
In the second section, we discuss and comment on the statistical and procedural techniques designed to attenuate or to avoid the problem. The statistical techniques are Harman's one-factor test, partial correlation procedure, and multiple method factors, among others. The procedural remedies include scale item trimming, temporal, proximal, psychological, or methodological separation of measurement, and protecting respondent anonymity. The procedural techniques are essentially related to research design, while the statistical ones are post hoc actions taken after data collection. It is obvious that the former is much more effective than the latter.
In the section that follows, we present the results of an extensive review and a comparison we conducted involving a total of 1596 papers in four prestigious Chinese journals issued in Taiwan during 1998-2003 and other four major journals published in English of the same time period. The Chinese periodicals selected are Journal of Management, Management Review, NSYSU Management Review, and NTU Management Review, while those in English are Administrative Science Quarterly, Academy of Management Journal, Journal of Management, and Journal of Organizational Behavior. Our focus was empirical studies that are quantitative in nature. Therefore, non-empirical research and empirical studies that are qualitative were not included. Furthermore, we decided not to examine studies adopting experiment methods because such techniques may be effective in avoiding CMV problems as independent variables are deliberately controlled.
Our investigation yielded interesting findings. Of the 871 studies in English during 1998-2003, there are 60.74% of them without CMV, 35.13% with partial CMV, and only 4.13% plagued with CMV. In contrast, of the 237 Chinese papers reviewed, the percentages of those without, with partial, and with CMV are 7.59%, 8.44%, and 83.97%, respectively. These results suggest that the majority of the Chinese papers did not handle or even recognize the CMV problem appropriately.
In the conclusion, we contend that a sound survey design is much better than 10 fancy statistical remedies and that a careful experimental design should effectively handle the CMV problems. There is room for improvement for our management community to do research without CMV and, eventually, to accumulate our knowledge more accurately.
英文關鍵字
Common Method Variance, Research Method, Construct Validity, Same Source Bias.
