I have performed a meta-analysis using five micro-array datasets. After performing meta analysis I visualized the heterogeneity using funnel plot and forest plot (using two up-regulated and two down-regulated genes). From the plots it can be concluded that for 3 genes (two top up and one top down), dataset 2 and 5 are outliers and for the remaining 1 gene (one top down), dataset 1 and 2 are outliers. For further analysis we are not sure whether to exclude the outliers or not. Also, if we decide to remove the datasets which datasets should we choose since three genes showed same pattern and one showed different? What method will be statistically accurate for deciding to exclude the datasets? Either by performing meta analysis with more datasets or by visualizing more genes via forest and funnel plots? I tried plotting funnel plots through the metafor package, to identify the heterogeneity values and outliers. After removing the 2nd dataset which was a common outlier, there was no significant reduction in heterogeneity (I square value). So should we instead remove all of the outliers, or keep them?
