This paper reports on a study of 3rd-grade students' modeling with data, which involves comprehensive investigations that draw upon STEM-based concepts, contexts, and questions, and generate products supported by evidence and open to informal inferential thinking. Within a real-world STEM-based context of licorice manufacturing, students experienced the "creation of variation" as they compared and represented the masses of "licorice sticks" they made by hand (using Play-Doh) and those using a Play-Doh extruder. By generating their own statistical measures, students could observe the features of data distributions including center, range, typical, and middle, at a much younger age than usual. They could draw inferences from the models they created, with awareness of how variation limits the certainty with which predictions can be made. The study supports a potential route for advancing early statistical learning. [For complete proceedings, see ED581294.]