Evaluating the in-Sample Prediction Approach to Assessing Class-Wide Impact for Class Certification

The “in-sample prediction approach” has been used in numerous recent antitrust class actions to assess whether all or almost all class members were harmed. The approach claims to estimate the impact of the challenged conduct on each transaction (e.g., estimate a distinct overcharge for each transaction in a price-fixing case). In every case to date, the approach found that almost all class members were harmed, and the courts certified the class. This article provides an overview of the approach and explains that it is unreliable because it fails two fundamental tests of reliable econometric methods. First, it is not a consistent estimator, meaning that it will not converge to the correct answer as the data grows large. Second, it has a false positive error rate far beyond typical thresholds, meaning that it will find that a substantial percentage of class members were harmed even if they were not harmed. Additionally, the approach is contradicted by the large academic literature on causal inference, which concludes that it is not generally possible to do what the approach claims to do (i.e., estimate the impact on each transaction).