Previous research has invested much effort in understanding how programming can contribute to the development of young learners' computational thinking (CT) in traditional K-12 classroom settings. Relatively few studies have examined programming for CT in informal online communities, especially for large scale quantitative research. With the repurposing and appropriation of code becoming standard practice for youth programmers in new and networked communities, this study proposes a model to investigate and explain how such remixing practices influence CT in learners. In particular, we use social cognitive theory to transform log data from more than 100,000 students and 200,000 programs in the Scratch online community into cognitive, environmental and behavioral factors. Then path modeling was used to show how the community environment can affect learners' remixing and further impact CT skills. The results demonstrate the intertwined relationships among the factors, generally showing that remixing can lead to CT improvement, but that over exposure to the community and excessive remixing can actually hinder CT development. Results and implications are further discussed.