A statistical approach to combining information from multiple sources in which it is assumed that the investigated relationship is constant across sources and any differences in individual results are attributable entirely to random variation.