As pointed out by Ferreira & Busatto, one parameter critically influencing the results of a coordinate-based meta-analysis is the FWHM of the kernel. The optimal FWHM has been found to depend on the meta-analytic method. ^{Reference Salimi-Khorshidi, Smith, Keltner, Wager and Nichols1} In signed differential mapping (SDM), a 25 mm FWHM shows a good compromise between sensitivity and control of false positives. ^{Reference Radua and Mataix-Cols2} This FWHM may account for different sources of spatial error such as registration mismatch, the size of original clusters or the location of the peak coordinates within the clusters. Much smaller FWHMs are common in activation/anatomical likelihood estimation (ALE), usually 10–15 mm. ^{Reference Rotge, Langbour, Guehl, Bioulac, Jaafari and Allard3} However, the use of these small FWHMs has not been clearly justified and it might lead to a dramatic reduction in sensitivity. Salimi-Khorshidi et al ^{Reference Salimi-Khorshidi, Smith, Keltner, Wager and Nichols1} found that the sensitivity of the ALE method with a standard deviation of 5 mm (corresponding to 10–15 mm FWHM) was approximately 50% of the sensitivity achieved with a standard deviation of 15 mm (corresponding to 35 mm FWHM).

Other limitations of ALE may be more serious ^{Reference Radua and Mataix-Cols2,Reference Laird, Fox, Price, Glahn, Uecker and Lancaster4} and have motivated the development of other methods such as SDM. ^{Reference Radua and Mataix-Cols2} For example, coordinates of increased and decreased activation (or, in this case, grey matter volume) are computed separately. This means that when calculating the meta-analytic increase in a voxel, the (negative) values of those studies reporting decreases in the same voxel are artificially replaced by zeros, leading to an inflation of the meta-analytic increase. Similarly, when computing the meta-analytic decrease, the (positive) values of those studies reporting increases in the same voxel are artificially replaced by zeros, leading to an inflation of the meta-analytic decrease. Therefore, brain regions with higher variability are more likely to be detected as significant in the meta-analysis, to the extent that some brain regions may appear to have both increases and decreases at the same time (e.g. see Menzies et al ^{Reference Menzies, Chamberlain, Laird, Thelen, Sahakian and Bullmore5} ). This is both mathematically and physiologically implausible. Another advantage of SDM is the strict inclusion of coordinates that are statistically significant at the whole-brain level and using the same threshold throughout the brain. ^{Reference Radua and Mataix-Cols2} This is of utmost importance given that it is not uncommon in neuroimaging studies that some regions (e.g. a priori regions of interest) are more liberally thresholded than the rest of the brain, thus potentially leading to false positives.

Unfortunately, psychiatric neuroimaging is plagued with methodological problems such as small sample sizes and overly liberal statistical methods, often making findings hard to replicate. Meta-analytical methods have the potential to overcome some of these limitations by helping researchers ‘see the forest before the trees’. However, if the methods or its parameters are not chosen rigorously, meta-analyses may suffer from the same problems that motivated their development in the first place.