“…Next, they individually coded the entire data set, double-coding approximately 10% to ensure that reliability was maintained in the process (interrater reliability remained above 80%). In all, 2,868 speech actions were identified and coded, with 123 turns marked with multiple codes (see Wilson, Edgington, Sztajn, & DeCuir-Gunby, 2014, for an analysis of all speech actions). For this study, we use a reduced data set of 322 speech actions about the four categories of interest that occurred during the 21 selected professional learning tasks targeting students' mathematical activity (see Appendix B online).…”