Activation likelihood estimation (ALE)

The wiki’s first coordinate-based meta-analysis method: a way of asking whether many independent neuroimaging studies — each with its own sample, task and disorder — nonetheless point at the same piece of cortex more often than chance would allow. It works on published peak coordinates rather than raw images, which is what lets it pool a heterogeneous literature the field could never re-scan.

The logic

Each reported activation peak (“focus”) is treated as an estimate of a true location blurred by measurement uncertainty, so it is modelled as a 3D Gaussian centred on the coordinate. The kernel width is scaled by sample size — larger studies get tighter, more confident kernels — implementing a random-effects inference over experiments. The per-study maps are unioned into a voxelwise ALE score, and that score is tested against a null distribution built by scattering the same number of foci randomly through a grey-matter mask. Voxels where the observed convergence beats the null survive; cluster-level FWE correction then ensures a surviving cluster is not the work of a single experiment.

The revised algorithm (Eickhoff et al. 2009, 2012; Turkeltaub et al. 2012) fixed earlier versions’ tendency to be driven by within-study and within-group effects. A simulation of 120,000 ALE meta-analyses (Eickhoff et al. 2016) established the rule of thumb the interoception literature now cites: ≥17 experiments are needed for cluster-level thresholding to control the excessive influence of any one study.

What it can and cannot claim

ALE answers “where do these studies converge?” and nothing about how much. Because no effect size enters — only the coordinates — it cannot be meta-regressed on moderators like age or sex, and a genuine effect that was under-powered in the source studies is as likely to be missed (false negative) as a spurious coordinate is to be included. It is a test of spatial agreement across a literature, with all of that literature’s selection and reporting biases baked in. Conjunction/contrast analyses extend it: intersecting two thresholded ALE maps asks whether two bodies of work (say, disrupted interoception vs. core affect) occupy the same voxels or different ones.

In the wiki

The method arrives with Nord, Lawson & Dalgleish (2021), which used it to ask whether disrupted interoceptive activation across psychiatric disorders converges anywhere, and found a single cluster in the left dorsal mid-insula, then used conjunction analyses to show that cluster is distinct from affect circuitry and from treatment-effect maps. The affect database Nord contrasted against — Lindquist et al. (2016) — is itself an ALE-family meta-analysis, and the wiki’s Lindquist et al. (2012) brain-basis-of-emotion paper is the same tradition applied to the locationist-vs-constructionist question. It sits beside lesion-symptom-mapping as the wiki’s other whole-literature synthesis method — one convergent-correlational, the other necessity-testing — and both are read for what a body of studies shows rather than what any one does.