Most studies linking features in brain imaging to traits such as cognitive abilities are too small to be reliable, argues a controversial analysis.