Models Mediation and Moderation

¶ … Mediation and Moderation

Models, Mediation, and Moderation

Critique of the approach to management and testing of mediator and moderator effects presented in Frazier et al. (2004) article

Frazier et al. (2004)'s article have introduced mediation analysis. After reading the article and comparing it with related ones, I have to raise concerns with about how Frazier et al. (2004) have used his method. I have reviewed recent and past methodological literature and developed a few recommendations on how Frazier et al. (2004) should have addressed three major issues. The first issue is the assumption of omitted variables, temporal order, and reliability. The authors have made brief visits to topics regarding confirmatory exploratory distinction and reliability. Additionally, in the provision of a feeling of the level at which previous literature studies were being practiced, Frazier et al. (2004) examined a sample comprising of 50 articles. Each of these articles had a minimum of analysis of mediation through the regression of ordinary least squares. A significant amount of the articles used exhibit problematic reporting; it appears that there is a room for improvement in carrying out this mediation analysis. Future studies and literature must consider demonstrating the extent at which improvements have been made to this situation (Preacher & Hayes, 2008).

In this article, Frazier et al. (2004) have posed a method commonly referred to as mediation analysis to demonstrate that a set of data is in line with a model whereby an intervening variable demonstrates how an independent variable influences a dependent variable. Studies of investing mediation have focused on understanding the temporal chain of events, which demonstrates how one event influences another. For example, how work environments influence judgments (Baron & Kenny, 1986). In this article, Frazier et al. (2004) presents a sample method based on regression that requires no specialized software that has made great influence. Nevertheless, there are empirical and theoretical reasons for raising concern on how Frazier et al. (2004) have applied this method in assessing mediation. The article focuses on the differences between mediation and moderation and with no explanation: it has not included extensive discussions regarding the complexities of structural equation modeling (SEM) and path modeling; these make any mediation analysis be perceived as a special case. Therefore, it is obvious that in Frazier et al. (2004) mediation analysis survey through SEM in his article published in counseling psychology journal, reported problems (Coe, 2000).

Frazier et al. (2004)'s article has a primary purpose of complementing previous literature through making explorations into a series of topics, which have been given a relatively little attention. Arrays of articles about the use of mediation analysis focusing on psychiatric and psychological studies have been published. However, Frazier et al. (2004) have not clarified how long mediation analysis should take in order to influence practice. It appears that we should expect some lag from the process and the article, as well. Therefore, for every topic that Frazier et al. (2004) have addressed some mediation analysis-based literature should document samples of mediation analysis in peer-reviewed journals. I would make recommendations of summarizing previous and current literature, which are likely to be useful in research studies intending to focus on mediation analysis. This should also include researchers interested in exploring future surveys of mediation analysis literature (Frazier, Tix & Barron, 2004).

I can criticize the significant non-significant Frazier et al. (2004) method, although I would promote the redundancy method at the same time. I have not favored the reflective inclusion of the Sobel test, as per the recommendations of ab'test and Holmbeck. I am in support of the application of additional tests based on the utility for a confidential interval to be created or on Type-I error and power rate considerations. In a study carried out on simulation of three categories of indirect influence tests, Frazier et al. (2004) established that no single class of test or one test examined under all other sets of tests is superior to others. Sobel test has developed into becoming less influential than the joint substantial test. Therefore, although it might be useful to apply additional tests, once assessments have been made to the joint significant of b and a, thus, treating ab' test as more definitive of all would not make any sense across the board (Coe, 2000).

I have observed weaker results in the ab' test sample. This is likely to be caused by either the excessiveness of Type-I error or the influential power of the integrated b and a test under some conditions. A bootstrap approach method would be a great recommendation when dealing with sample sizes that are less than 20 (Preacher & Hayes, 2008). I have established this recommendation based on performance comparisons of three single sample approaches and a program of SAS in the application of five methods availed by RTI international. I suggest that Frazier et al. (2004) consult Mackinnon to establish which approach is likely to have the best properties of statistics for his data and engage any of the available programs to fulfill his desires of a resampling method. In conclusion, I recommend Frazier et al. (2004) to ensure caution while using the partial correlation approach. In some circumstances, partial correlation is likely to have desirable statistical properties: it has a distinctively null hypothesis, compared to other intervening variable tests. This means that the test based on partial correlation does not make direct assessments on the significance of the purported indirect and direct influence (Baron & Kenny, 1986).

The coefficients provide descriptive information, which is the center of conceptual issues. Values of coefficients can be applied to make judgments on the rate of data consistency with hypothesized relations. Coefficient values can be engaged to evaluate the practical benefits of the findings or even describe the power of observed relations. The results of significant tests cannot purely convey this information. Although data is inconsistent with mediation due to the power of non-significant tests, assessing the coefficients of b and a' test allows researchers to assess if the relation of a proposed mediator to a treatment is weak than it has been hypothesized. When mediation of data support via significant methods of mediation tests, the level of indirect influence is crucial as the estimated product of b and a' test reflect (Frazier, Tix & Barron, 2004).