US courts are demanding greater economic rigor in analyses provided by both plaintiffs and defendants at the class certification stage. In this paper, the authors examine a selection of court decisions on class certification since Comcast Corp. v. Behrend with a focus on a particular subset of specification testing that examines the propensity for the plaintiffs’ model, unaltered, to yield faulty predictions. A faulty prediction can arise when an econometric model finds impact or overcharges in populations where the plaintiffs’ theory of harm would predict that no impact should exist, or finds no effect where the allegations indicate one should be present. Our examination of the decisions post-Comcast suggests that courts may be most persuaded that a model is unreliable for showing impact and damages at trial if the defendant can empirically demonstrate that the model as constructed yields faulty predictions. On the other hand, the court is more likely to find the plaintiffs’ model sufficient for the purpose of class certification when the defendant uses an alteration of the plaintiffs’ model to show its propensity towards generating faulty predictions, particularly when case-specific facts and economic theory do not provide ample justification for such alterations.