关键词: 共同方法变异 共同方法偏差 Harman单因子法 ULMC法 CFA标签变量法

Abstract: Common method bias (CMB) test is a routine step in empirical studies of psychology since the same measurement method probably produces common method variance (CMV) in variables that might falsely inflate or deflate observed relationships among measures. There are three popular statistical approaches for testing CMB including Harman’s single-factor test, controlling for the effects of an unmeasured latent methods factor (ULMC) technique and confirmatory factor analysis (CFA) marker technique. First, to better understand the effects of CMV, we analyzed the mathematical model of CMV to illustrate how it inflated or deflated the observed relationships. Obviously, the inflation or deflation of the observed relationships caused by CMV were not only associated with the method and traits themselves, but also related to the correlations between the method and traits. Then, we discussed the statistical power of Harman’s single-factor test, ULMC, and CFA marker technique for testing CMB. Harman’s single-factor test showed the lowest power, and ULMC technique had median power in CMB test. Although CFA marker technique had the highest power, there were also some shortcomings such as the difficulties in finding an ideal marker variable. Many researchers claimed that their marker variables were ideal, but the fact was that only about half of the researches adopted ideal marker variables. A simulated sample with 40% CMV was tested by the three approaches mentioned above. As a result, only CFA marker technique indicated that serious CMB existed in the sample. Based on the previous analyses, we proposed that a good approach to testing CMB should have at least three features: reflecting the mathematical model of CMV, stability in CMV test, and sensitiveness to the changes of CMV and CMB. The mathematical model of CMV means that observed variables’ variances are only from three sources: method, traits, and random errors, the CMV may cause changes in correlations between different traits. Therefore, any statistical approach for testing CMB should reflect these three kinds of variances, especially the method variances. But as for CFA marker technique which can extract most of the method variance, marker variable only represents method to some extent. With regard of the stability, a good evaluation criterion will not be affected by factors other than CMV. For example, Harman’s single-factor test is serious affected by questionnaire reliability and the number of traits. The sensitiveness means that a good approach sensitively responds the change of CMV and CMB. The previous simulation study found that ULMC technique, one of the most popular statistical approaches was less sensitive to the change. In conclusion, Harman’s single-factor test is not a good approach to test CMB, and ULMC technique is better. Although CFA marker technique has relatively high statistical power, it is too complicated and its procedures for testing CMB are tedious. In order to simplify the steps, the test of Model C can be deleted given that the actual data is usually difficult to meet the assumption of Model C. Till now, we don’t have an ideal approach to test CMB yet. Even if we would have, the best way to minimize the CMV is through the beforehand control.

Key words: common method variance, common method bias, Harman’s single-factor test, ULMC Technique, CFA marker technique