Research School of Population Health, Australian National University, Canberra, Australia.
Research School of Population Health, Australian National University, Canberra, Australia
Perspect Psychol Sci. 2016 May;11(3):408-14. doi: 10.1177/1745691616635599.
Ferguson's (2015a) meta-analysis assessed a very important and controversial topic about children's mental health and video games. In response to the concerns raised by researchers about the appropriateness of the meta-analytical methods used by Ferguson; we decided to reanalyze the data and discuss two major misconceptions about meta-analysis. We argue that partial correlations can (and should) be meta-analyzed instead of zero-order bivariate correlations if the predictors included in the partial correlation represent a similar construct. We also discuss the fallacy by which the conventional meta-analytical model assumes that the studies' effect sizes came into being according to the same random effect construct used by the analysis. Our replication results using partial correlations, standardized (valid and reliable) outcomes, and an improved meta-analytical model (that does not assume a random effect is the mechanism of data generation) confirmed the main results of Ferguson's meta-analysis. There was a significant yet very small effect on aggressive behavior of exposure to both general, rp = 0.062, 95% CI [0.012, 0.112], and violent, rp = 0.055, 95% CI [0.019, 0.091], video games. A very small effect was seen on reduced prosocial behavior, but this was only in the general video game exposure category, rp = 0.072, 95% CI [0.045, 0.100].
弗格森（2015a）的荟萃分析评估了一个非常重要且有争议的关于儿童心理健康和视频游戏的话题。针对研究人员对弗格森使用的元分析方法的适当性提出的担忧；我们决定重新分析数据，并讨论两个关于元分析的主要误解。我们认为，如果偏相关中包含的预测因素代表相似的结构，那么可以（并且应该）对偏相关进行元分析，而不是对零阶双变量相关进行元分析。我们还讨论了传统元分析模型的谬误，即分析中使用的相同随机效应结构产生了研究的效应大小。我们使用偏相关、标准化（有效和可靠）结果和改进的元分析模型（不假设随机效应是数据生成的机制）进行的复制结果证实了弗格森元分析的主要结果。接触一般暴力和暴力视频游戏对攻击行为有显著但非常小的影响，rp = 0.062，95%CI [0.012，0.112]，rp = 0.055，95%CI [0.019，0.091]。减少亲社会行为也有非常小的影响，但这只出现在一般视频游戏暴露类别中，rp = 0.072，95%CI [0.045，0.100]。
