Corrigendum to “Marginal Effects in Interaction Models: Determining and Controlling the False Positive Rate”

After publication of Esarey and Sumner, we recognized substantively significant errors in the paper. We correct those errors here. We discovered these errors as part of work on another, unrelated paper to which one co-author hoped to apply similar methods. The Brambor et al. procedure is overconfident when separately testing multiple hypotheses, as correctly stated in our paper. However, when conjointly testing multiple hypotheses, the Brambor et al. procedure is appropriate. The procedure we suggest on pp. 1161–1163 to correct for overconfidence in the case of separate testing of multiple hypothesis (based on Benjamini and Hochberg) is (to our knowledge) correct, but subject to several limitations unstated in the paper. This corrigendum lays out all errors and limitations and adds a more robust procedure to our interactionTest software.